JP2007048006A - Image processor and image processing program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To easily determine the recognition result of the object of observation according to the level of the success/failure of recognition, and to perform processing based on the recognition result according to the level of the success/failure of recognition. <P>SOLUTION: This image processor 1 is provided with a cell recognizing part 3 for recognizing an object region corresponding to the object of observation from an observation image, a reference edge image generating part 4 for generating a reference edge image by detecting an edge included in the observation image, a recognition label image separation part 5a for separating the outline part of the object region and the inside of the region, an outline part edge component calculating part 5b for calculating an outline part edge component strength average as a mean value of pixel values in the reference edge image of the outline part of the object region, an inside edge component calculating part 5c for calculating an internal edge component intensity average as the mean value of the pixel values in the reference edge image in the object region and a recognition reliability deciding part 5d for associating recognition reliability showing the level of the success/failure of the recognition for each object region based on the outline part edge component strength average and the internal edge component strength average. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an image processing apparatus and an image processing program for recognizing an image region corresponding to an observation target from input images.

  Conventionally, observation using various cells as specimens using a microscope or the like has been performed. For example, when observing live cells over time using a microscope, individual live cells are usually extracted from each observation image obtained by imaging live cells at multiple time points by visual and manual operations. The time-dependent change in the feature amount of each live cell is observed by comparing between observation images at different observation points, and each live cell is correlated and tracked between the observation images.

  Such a conventional observation method has a problem that the load on the observer increases as the number of observation cells, observation frequency, observation range, observation time, etc. increase. In contrast, a technique has been proposed in which cell regions corresponding to individual cells are automatically extracted from the observed image by image processing to reduce the load on the observer (for example, see Patent Document 1). .

  In the cell shape extraction method disclosed in Patent Document 1, a cell shape as a model is applied to a cell in an observation image to detect a cell center position, and a concentration gradient around the detected cell center position is obtained. The degree of coincidence between the density gradient direction and the direction from each pixel toward the cell center position is calculated to extract the outline of the cell region.

  More specifically, first, a calculation process that calculates a density gradient direction at an arbitrary pixel in the observation image and gives a score to a pixel separated from the arbitrary pixel in the density gradient direction by a predetermined radius is performed in the observation image. Repeat for all pixels. As a result of this processing, the pixel corresponding to the cell center obtains a high score by concentrating the scores from each pixel. Next, since the concentration gradient direction in the cell contour portion is directed toward the cell center, the contour of the cell region is determined based on the inner product of the concentration gradient vector of the surrounding pixels at the cell center and the vector from the arbitrary pixel toward the cell center. Candidate pixels that are candidates for the contour pixel to be shown are extracted. Thereafter, based on the extracted candidate pixels, a detailed outline of the cell region is extracted using a dynamic outline extraction method. In the active contour extraction method, a circular closed curve having a specified radius centered on the cell center position is set, and when the set closed curve overlaps with a candidate pixel, the density gradient of the candidate pixel is reduced. Then, the evaluation value is calculated while locally changing the shape of the closed curve using the sum of the magnitudes of the concentration gradients of the closed curve as an evaluation value, and the closed curve when the evaluation value is minimized is defined as the final contour of the cell region. To do.

  As another area extraction method, there is a method using an image processing technique called watershed area division (see, for example, Non-Patent Document 1). In this region extraction method, the image is divided for each area where low-brightness pixels are concentrated, but since cells generally appear as a set of high-brightness pixels, the observation image is divided for each area where high-brightness pixels are concentrated. Thus, the cell region is extracted. For this reason, when extracting a cell area | region, it applies as what inverts brightness | luminance and divides a high-intensity area | region on a technique.

Japanese Patent No. 3314759 Vincent & Soille, "Watersheds in Digital Spaces: An Efficient Algorithm Based on Immersion Simulations", IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL.13, NO.6, JUNE 1991

  By the way, in the automatic extraction of a cell region by image processing, it is generally difficult to obtain a complete result, and an erroneous extraction result often occurs. For example, in the cell shape extraction method disclosed in Patent Document 1, since the outline of a cell region is extracted based on the cell shape as a model, it is difficult to capture the cell shape deviating from the model. . In addition, for example, when observing a fluorescent image obtained by imaging a living cell on which a fluorescent protein is allowed to act and extracting a cell region corresponding to the living cell, due to the influence of noise and fluctuations in the amount of fluorescence, The cell region may not be detected stably. When an error occurs in the extraction of the cell region, there is a problem that the error propagates to processing such as measurement of the feature amount and tracking of the living cell performed based on the extraction result.

  The present invention has been made in view of the above, and in a recognition process for automatically recognizing a region corresponding to an observation object from an observation image, an evaluation result obtained by evaluating the degree of success or failure of the recognition is correlated. A recognition result can be generated. As a result, the recognition result can be easily identified according to the degree of recognition success, and the processing based on the recognition result can be processed according to the degree of recognition success. It is an object to provide a processing device and an image processing program.

  In order to achieve the above object, an image processing apparatus according to claim 1 is an image processing apparatus that performs recognition processing for recognizing a target area that is an image area corresponding to an observation target from input observation images. A recognition reliability determination unit that determines the degree of success or failure of recognition by the recognition process for each target region is provided.

  In the image processing apparatus according to claim 2, in the above invention, the recognition reliability determination unit uses the edge pixel value of each pixel of the edge image in which the edge included in the observation image is detected. It is characterized by determining the degree of success or failure of recognition by processing.

  In the image processing apparatus according to claim 3, in the above invention, the recognition reliability determination unit includes an edge pixel value included in a pixel in the edge image corresponding to a contour pixel included in a contour portion of the target region. The recognition processing using the edge strength of the edge and the edge strength in the region that is the edge pixel value of the pixel in the edge image corresponding to the inner pixel included in the region excluding the contour portion of the target region. It is characterized in that the degree of success or failure of recognition is determined.

  According to a fourth aspect of the present invention, in the above invention, a temporary area recognition unit for recognizing a temporary target area from the observed image, and a temporary target area recognized by the temporary area recognition unit. From the region integration means for forming the integrated region by integrating the adjacent temporary target regions based on the feature amount indicating the feature between the adjacent temporary target regions, and between the integrated temporary target regions in the integrated region Selective area recognition means for recognizing the temporary target area or the integrated area as the target area according to an edge pixel value of a pixel in the edge image corresponding to the boundary pixel included in the boundary line of It is characterized by having.

  According to a fifth aspect of the present invention, in the above invention, a temporary area recognition unit that recognizes a temporary target area from the observed image, and a temporary target area recognized by the temporary area recognition unit. From the region integration means for forming the integrated region by integrating the adjacent temporary target regions based on the feature amount indicating the feature between the adjacent temporary target regions, and between the integrated temporary target regions in the integrated region Selective area recognition means for recognizing the temporary target area or the integrated area as the target area according to an edge pixel value of a pixel in the edge image corresponding to the boundary pixel included in the boundary line of The recognition reliability determination means uses the edge pixel value of the pixel in the edge image corresponding to the boundary pixel as the in-region edge strength, and the degree of success or failure of recognition by the recognition process And judging.

  In the image processing apparatus according to claim 6, in the above invention, the recognition reliability determination unit uses the matching degree between the target area recognized by the recognition process and a preset area model. It is characterized by determining the degree of success or failure of recognition by the recognition process.

  According to a seventh aspect of the present invention, in the above invention, the region model is defined based on the shape or luminance distribution of the observation target.

  According to an eighth aspect of the present invention, there is provided the image processing apparatus according to the above invention, wherein the display control means performs control to display each target region recognized by the recognition processing in a mode according to the determination result of the recognition reliability determination means. Is further provided.

  In the image processing apparatus according to claim 9, in the above invention, the observation image includes a plurality of images taken at different observation time points, and the observation object is identical between the observation images at different observation time points. The object tracking means further includes a target tracking means for reflecting the degree of success or failure of recognition by the recognition processing determined by the recognition reliability determination means in the determination of the identity of the observation target. And

  According to a tenth aspect of the present invention, there is provided the image processing apparatus according to the above invention, wherein the recognition success level is low in correspondence with the target region that is determined to have a low recognition success level by the recognition reliability determination unit. The information processing device further includes notification control means for performing control for notifying warning information.

  The image processing apparatus according to claim 11 is characterized in that, in the above invention, the observation object is a cell.

  According to a twelfth aspect of the present invention, there is provided an image processing program for causing an image processing apparatus that performs recognition processing for recognizing a target region that is an image region corresponding to an observation target from the input observation images to perform the recognition processing. In the image processing program, the image processing apparatus is caused to execute a recognition reliability determination procedure for determining the degree of success or failure of recognition by the recognition processing for each target region.

  In the image processing program according to claim 13, in the above invention, the recognition reliability determination procedure uses the edge pixel value of each pixel of the edge image in which an edge included in the observation image is detected. It is characterized by determining the degree of success or failure of recognition by processing.

  In the image processing program according to claim 14, in the above invention, the recognition reliability determination procedure includes an edge pixel value included in a pixel in the edge image corresponding to a contour pixel included in a contour portion of the target region. The recognition processing using the edge strength of the edge and the edge strength in the region that is the edge pixel value of the pixel in the edge image corresponding to the inner pixel included in the region excluding the contour portion of the target region. It is characterized in that the degree of success or failure of recognition is determined.

  According to a fifteenth aspect of the present invention, there is provided an image processing program according to the above invention, wherein a temporary region recognition procedure for recognizing a temporary target region from the observed image, and a temporary target region recognized by the temporary region recognition procedure. From the region integration procedure for forming the integrated region by integrating the adjacent temporary target regions based on the feature amount indicating the feature between the adjacent temporary target regions, and between the integrated temporary target regions in the integrated region A selective region recognition procedure for recognizing the temporary target region or the integrated region as the target region according to an edge pixel value of a pixel in the edge image corresponding to a boundary pixel included in the boundary line of It is made to perform.

  According to a sixteenth aspect of the present invention, there is provided an image processing program according to the above invention, wherein a temporary region recognition procedure for recognizing a temporary target region from the observed image, and a temporary target region recognized by the temporary region recognition procedure. From the region integration procedure for forming the integrated region by integrating the adjacent temporary target regions based on the feature amount indicating the feature between the adjacent temporary target regions, and between the integrated temporary target regions in the integrated region A selective region recognition procedure for recognizing the temporary target region or the integrated region as the target region according to an edge pixel value of a pixel in the edge image corresponding to a boundary pixel included in the boundary line of The recognition reliability determination procedure is executed to recognize an edge pixel value of a pixel in the edge image corresponding to the boundary pixel as the in-region edge strength by the recognition process. And judging a degree of success.

  The image processing program according to claim 17 is the image processing program according to the above invention, wherein the recognition reliability determination procedure uses the matching degree between the target area recognized by the recognition process and a preset area model. It is characterized by determining the degree of success or failure of recognition by the recognition process.

  The image processing program according to claim 18 is the above invention, wherein the observation image includes a plurality of images taken at different observation time points, and the observation object is identical between the observation images at different observation time points. The target tracking procedure is further executed, and the target tracking procedure reflects the success or failure of recognition by the recognition processing determined in the recognition reliability determination procedure in the determination of the identity of the observation target. Features.

  According to the image processing apparatus and the image processing program according to the present invention, in the recognition processing for automatically recognizing the region corresponding to the observation target from the observation image, the evaluation result that evaluates the degree of success or failure of the recognition is associated A recognition result can be generated. As a result, the recognition result can be easily identified according to the degree of success or failure of the recognition, and processing based on the recognition result can be processed according to the degree of success or failure of the recognition.

  Exemplary embodiments of an image processing apparatus and an image processing program according to the present invention are explained in detail below with reference to the accompanying drawings. In addition, this invention is not limited by this embodiment. In the description of the drawings, the same parts are denoted by the same reference numerals.

(Embodiment 1)
First, an image processing apparatus and an image processing program according to the first embodiment of the present invention will be described. FIG. 1 is a block diagram illustrating a configuration of the image processing apparatus 1 according to the first embodiment. FIG. 2 is a block diagram showing a detailed configuration of the image matching unit 5 shown in FIG.

  As illustrated in FIG. 1, the image processing apparatus 1 recognizes a cell region that is an image region corresponding to a cell from an image acquisition unit 2 that acquires an observation image in which a cell to be observed is recorded, and the observation image. Recognition by cell recognition processing using a cell recognition unit 3 that performs cell recognition processing, a reference edge image generation unit 4 that generates a reference edge image as an edge image in which an edge included in an observed image is detected, and a reference edge image Of the image collation unit 5 that determines the degree of success or failure for each cell region, a display unit 6 that displays various types of information such as image display, a storage unit 7 that stores various types of information such as images, and the like. And a control unit 8 that controls processing and operation of each unit. The image acquisition unit 2, cell recognition unit 3, reference edge image generation unit 4, image collation unit 5, display unit 6 and storage unit 7 are electrically connected to the control unit 8.

  The image acquisition unit 2 includes various interfaces such as a communication interface such as USB and IEEE1394, and an interface corresponding to a portable storage medium such as a flash memory, a CD, a DVD, and a hard disk, and an observation image is received from an external device via the interface. get. The image acquisition unit 2 includes an imaging device realized by using an imaging lens, an imaging element such as a CCD, and an A / D converter, and acquires an observation image generated by imaging by the imaging device. Also good. In this case, various types of imaging devices capable of generating digital images, such as a digital camera, a microscope, and a visual sensor, can be applied to the imaging device included in the image acquisition unit 2. The image acquisition unit 2 outputs the acquired observation image to the cell recognition unit 3 and the reference edge image generation unit 4. In addition, the image acquisition unit 2 can output the observation image to the storage unit 7 for storage.

  Here, as an example, the observation image acquired by the image acquisition unit 2 is a cell fluorescence image obtained by imaging cells in a biological tissue that has been stained with a fluorescent dye in advance. In such a cell fluorescence image, the corresponding part of the cell on which the dye acts is observed brightly. That is, the cell region in the cell fluorescence image exists as an image region where high luminance pixels are concentrated. The cell staining may be for staining the entire cell, or for staining only a specific site such as the cell nucleus, actin, or cell membrane. The dye used for staining is not limited to a fluorescent dye, and any dye may be used as long as it further enhances cell contrast and does not alter cell characteristics in addition to contrast. . That is, the cell image as the target image in which the cells to be observed are recorded is not limited to the cell fluorescence image, and may be various cell-stained images.

  The observation image acquired by the image acquisition unit 2 may be an arbitrary image as long as the observation target can be identified, such as a monochrome image, a color image, or a color difference image. Furthermore, the observation target recorded in the observation image need not be interpreted as being limited to cells, and may be any object such as a vehicle, a person, or an animal.

  The cell recognition unit 3 includes a smoothing unit 3a, a gradient direction calculation unit 3b, a reaching extreme value calculation unit 3c, a region integration unit 3d, a background removal unit 3e, and a labeling unit 3f. A cell region is extracted from the inside and recognized, and a recognition label image indicating the recognition result is generated.

  Here, the smoothing unit 3a smoothes the cell fluorescence image to reduce luminance unevenness in the image. The gradient direction calculation unit 3b calculates the gradient direction of the luminance (pixel value) of each pixel in the smoothed cell fluorescence image, and detects an extreme pixel indicating the maximum value of luminance. The gradient direction is defined as a direction from a pixel to be processed toward a pixel having the highest luminance value among pixels adjacent to the pixel to be processed. The ultimate value calculation unit 3c generates a temporary label image obtained by dividing the cell fluorescence image into temporary cell regions based on the gradient direction and the extreme pixel. The region integration unit 3d integrates adjacent temporary cell regions of the temporary label image so as to match the outline shape of the cells in the cell fluorescence image, and forms a cell region as a target region. The background removal unit 3e removes a pixel having a small luminance value in the cell fluorescence image from the cell region as a background pixel indicating the background. The labeling unit 3f generates a recognition label image in which a unique region label is assigned to each cell region.

  In addition, the cell recognition unit 3 outputs the generated recognition label image to the image matching unit 5. The cell recognizing unit 3 can also output and store the generated recognition label image to the storage unit 7 via the control unit 8. The cell recognition unit 3 can also generate a recognition label image by acquiring the cell fluorescence image stored in the storage unit 7.

  The reference edge image generation unit 4 includes a smoothing unit 4a, a structure preserving smoothing unit 4b, an edge extracting unit 4c, an outline emphasizing unit 4d, and a gradation converting unit 4e, and includes an edge included in the cell fluorescence image acquired from the image acquiring unit 2 To generate a reference edge image.

  Here, the smoothing unit 4a smoothes the cell fluorescence image to reduce luminance unevenness in the image. The structure preserving smoothing unit 4b leaves a valley of luminance values corresponding to the outline of the cell region in the cell fluorescence image smoothed by the smoothing unit 4a, and a valley of minute luminance values that are noise inside the cell region. Further smoothing is performed to reduce. The edge extraction unit 4c detects, from the cell fluorescence image smoothed by the structure preserving smoothing unit 4b, a luminance value change indicating the boundary between the cell region and the background or between the cell regions as an edge. Is generated. The contour emphasizing unit 4d weakens the edge component inside the cell region among the edges included in the edge image, and emphasizes the edge component of the contour portion of the cell region. The gradation converting unit 4e performs gradation conversion so as to weaken weak edges based on the edge image processed by the contour emphasizing unit 4d to generate a reference edge image.

  Further, the reference edge image generation unit 4 outputs the generated reference edge image to the image matching unit 5. Note that the reference edge image generation unit 4 can also output and store the generated reference edge image to the storage unit 7 via the control unit 8. The reference edge image generation unit 4 can also acquire a cell fluorescence image stored in the storage unit 7 and generate a reference edge image.

  As shown in FIG. 2, the image matching unit 5 includes a recognition label image separation unit 5a, a contour edge component calculation unit 5b, an internal edge component calculation unit 5c, and a recognition reliability determination unit 5d, and each pixel of the reference edge image Is associated with each cell region recognized by the cell recognition unit 3 using the edge pixel value of the.

  Here, the recognition label image separation unit 5a separates the contour part of the cell region from the inside of the region from the recognition label image generated by the cell recognition unit 3, and extracts the contour part label image obtained by extracting the contour part and the region inside. An extracted internal label image is generated. The contour portion edge component calculation unit 5b is a contour portion edge component that is an average value of edge pixel values of pixels in the reference edge image corresponding to the contour pixels included in the contour portion of the cell region indicated in the contour portion label image. Calculate the intensity average. The internal edge component calculation unit 5c calculates an internal edge component intensity average that is an average value of edge pixel values of pixels in the reference edge image corresponding to internal pixels included in the cell region indicated in the internal label image. . The recognition reliability determination unit 5d determines the degree of success or failure of the recognition for each cell region recognized by the cell recognition unit 3 based on the contour edge component intensity average and the internal edge component intensity average.

  Further, the image matching unit 5 outputs the recognition label image and the recognition reliability indicating the degree of success or failure of the recognition to the display unit 6. The image matching unit 5 can also output and store the recognition reliability to the storage unit 7 via the control unit 8. The image matching unit 5 can also acquire the recognition label image and the reference edge image stored in the storage unit 7 and determine the degree of success or failure of the recognition.

  The display unit 6 includes a display device having a CRT, a liquid crystal display, and the like, acquires the recognition label image and the recognition reliability output from the image matching unit 5, and displays them as image information and numerical information. In this display, the display unit 6 can display only one of image information and numerical information. Both can be displayed simultaneously or by switching. Note that the display unit 6 may acquire and display the recognition label image and the recognition reliability stored in the storage unit 7.

  The storage unit 7 is realized by using a ROM that stores various processing programs and the like, and a RAM that stores processing parameters, processing data, and the like for various processes. The storage unit 7 particularly recognizes a program for causing the cell recognition unit 3, the reference edge image generation unit 4 and the image collation unit 5 to perform processing, that is, recognition for recognizing a cell region from the cell fluorescence image by the image processing apparatus 1. An image processing program for executing the recognition reliability determination process for associating the recognition reliability with the process is stored. Note that the storage unit 7 may include a portable storage medium such as a flash memory, a CD, a DVD, or a hard disk as a removable storage unit.

  The control unit 8 is realized using a CPU or the like that executes various processing programs stored in the storage unit 7. In particular, the control unit 8 executes an image processing program stored in the storage unit 7 and controls processing and operations of the cell recognition unit 3, the reference edge image generation unit 4, and the image matching unit 5 according to the image processing program. . In addition, the control unit 8 performs control to display the recognition label image and the recognition reliability output from the image matching unit 5 on the display unit 6 as image information and numerical information. At this time, the display unit 6 is controlled so as to display each cell region recognized by the cell recognition unit 3 in a manner corresponding to the recognition reliability. In addition to the recognition label image and the recognition reliability, the control unit 8 controls the display unit 6 to display a cell fluorescence image as an original image, a smoothed cell fluorescence image, a temporary label image, a reference edge image, and the like. Can also be done.

  In addition, the control unit 8 performs control for notifying warning information indicating that the recognition success level is low for the cell region determined by the recognition reliability determination unit 5d as having low recognition success level. In other words, for the cell region with low recognition reliability, the observer is informed that there is a high possibility of misrecognition and that there is a high possibility that erroneous processing will be performed in various processes performed based on the recognition result. Control to notify.

  Here, a processing procedure performed by the image processing apparatus 1 will be described. FIG. 3 is a flowchart showing an image processing procedure in which the image processing apparatus 1 processes the cell fluorescence image by the control unit 8 executing the image processing program. As shown in FIG. 3, first, the image acquisition unit 2 acquires a cell fluorescence image from an external device (step S <b> 101), and the cell recognition unit 3 extracts a cell region from the cell fluorescence image acquired from the image acquisition unit 2. Recognition processing is performed to generate a recognition label image (step S103).

  Subsequently, the reference edge image generation unit 4 performs a reference edge image generation process of detecting an edge included in the cell fluorescence image acquired from the image acquisition unit 2 and generating a reference edge image (step S105). The image matching unit 5 uses the reference edge image acquired from the reference edge image generation unit 4 to perform a recognition reliability determination process for associating the recognition reliability with each cell region in the recognition label image acquired from the cell recognition unit 3. (Step S107). The display unit 6 displays the recognition label image acquired from the image matching unit 5 and the recognition reliability as a recognition result (step S109). Thereafter, the control unit 8 ends the series of image processing.

  Note that the control unit 8 can repeatedly perform the processing of steps S101 to S109 until, for example, instruction information for ending predetermined processing is received. Further, instead of acquiring the cell fluorescence image from the image acquisition unit 2 in step S101, the cell fluorescence image stored in the storage unit 7 can be acquired and the processing from step S103 can be executed. Furthermore, instead of processing steps S101 to S105, the recognition label image and the reference edge image stored in the storage unit 7 can be acquired and step S107 can be executed. Also, step S109 can be omitted. Note that the control unit 8 may perform step S103 and step S105 by switching the processing order.

  Next, the details of each process will be described with respect to steps S103 to S109 shown in FIG. First, the cell recognition process in step S103 will be described. FIG. 4 is a flowchart showing the procedure of the cell recognition process. As illustrated in FIG. 4, the smoothing unit 3 a performs a smoothing process for smoothing the cell fluorescence image acquired from the image acquisition unit 2 (step S <b> 111), and the gradient direction calculation unit 3 b includes the smoothed cell fluorescence image. The gradient direction calculation process for calculating the gradient direction of the luminance of each pixel and detecting the extreme value pixel is performed (step S113).

  Subsequently, the ultimate extreme value calculation unit 3c divides the cell fluorescence image into temporary cell regions based on the gradient direction and extreme value pixels acquired from the gradient direction calculation unit 3b, and generates a temporary label image. A value calculation process is performed (step S115). The region integration unit 3d performs region integration processing for integrating adjacent temporary cell regions in the temporary label image to form a cell region (step S117). The background removal unit 3e performs a background removal process of removing a pixel having a small luminance value in the cell fluorescence image from the cell region as a background pixel (step S119). The labeling unit 3f performs a labeling process for generating a recognition label image by assigning a unique region label to each shaped cell region (step S121). Then, the control unit 8 causes the image collation unit 5 to output the recognition label image from the labeling unit 3f, and then proceeds to the process of step S105.

  In step S111, the smoothing unit 3a smoothes the cell fluorescence image while reducing the luminance unevenness while preserving the structure of the luminance distribution showing a large luminance value change such as an edge in the cell fluorescence image. The smoothing unit 3a can output and store a smoothed image as a result of the smoothing process to the storage unit 7 via the control unit 8.

  In step S113, the gradient direction calculation unit 3b first calculates the luminance value of the processing target pixel from the luminance value of the adjacent pixel between the processing target pixel and each of the eight neighboring pixels centered on the processing target pixel. The subtracted luminance value difference is calculated, and the direction from the processing target pixel toward the adjacent pixel having the maximum luminance value difference is defined as the gradient direction. Then, this gradient direction is recorded in association with the processing target pixel. Further, the gradient direction calculation unit 3b determines that the pixel to be processed is an extreme pixel indicating a maximum value in the vicinity of 8 when the luminance value differences calculated between the adjacent pixels in the vicinity of 8 are all negative values. Record. The gradient direction calculation unit 3b sequentially performs such processing on all the pixels in the cell fluorescence image.

  In step S115, the reached extreme value calculation unit 3c sequentially follows the gradient direction from the processing target pixel until it reaches the extreme value pixel, and associates each pixel that has become a path with the reached extreme value pixel. This process is sequentially performed on all the pixels in the cell fluorescence image, and each pixel is associated with one of the extreme pixels. Then, the ultimate extreme value calculation unit 3c recognizes each region of each extreme value pixel as a temporary cell region, assuming that each pixel associated with the same extreme value pixel constitutes the same region, A temporary label image is generated by assigning a unique region label to the cell region. In the temporary label image generated in this way, since the temporary cell region is formed for each extreme pixel corresponding to a minute change in the luminance value in the cell fluorescence image, it normally exists in the cell fluorescence image. It is divided into temporary cell regions that are finer than the cell region. In addition, if the area | region label | marker provided to a temporary cell area | region is intrinsic | native, arbitrary description using a number, an alphabet, a symbol, etc. may be sufficient.

  In step S117, the region integration unit 3d, for example, integrates the adjacent temporary cell regions and sets the cell region when the distance between the extreme values between the adjacent extreme cell regions is equal to or less than a predetermined threshold. To do. Here, the distance between extreme values is a distance along the luminance distribution curve indicating the distribution of luminance values of the cell fluorescence image, as shown in FIG. 5, and is a luminance distribution curve corresponding to each of different extreme pixels. The distance connecting the top points. In FIG. 5, the length of the partial curve shown by the bold line connecting two points on the luminance distribution curve is the distance between extreme values.

  Further, the region integration unit 3d integrates adjacent temporary cell regions by, for example, comparing the edge image and the temporary label image as generating an edge image in which an edge included in the cell fluorescence image is detected. You can also. In this case, as shown in FIG. 6, the region integration unit 3d has an intensity value of a pixel in the edge image corresponding to a pixel indicating a boundary line between adjacent temporary cell regions in the temporary label image equal to or less than a predetermined threshold value. In this case, the pixels indicating the boundary line are removed, and adjacent temporary cell regions are integrated. In FIG. 6, the boundary line indicated by the arrow in the temporary label image is weak and hardly recognized in the edge image, and is thus removed as being different from the outline of the cell region.

  The region integration unit 3d performs the region integration processing on all adjacent temporary cell regions existing in the temporary label image, and repeats until the integration cannot be performed. And the temporary cell area | region left as a result of repeating integration is recognized as a final cell area | region. In addition, the temporary cell area | region which was not integrated by the area | region integration part 3d is recognized as a cell area as it is.

  In step S119, the background removal unit 3e uses, as background pixels, all pixels included in the cell region corresponding to, for example, a pixel having a luminance value equal to or lower than a predetermined threshold or an extreme value pixel having a luminance value equal to or lower than a predetermined threshold. Remove.

  In step S121, the labeling unit 3f may newly give a unique area label for each cell area, or may take over the area label given to the temporary cell area. In the case of taking over the region label, the labeling unit 3f takes over any region label of each temporary cell region before the integration into the cell region as a result of the integration of the temporary cell region. In addition, the area | region label | marker provided to a cell area | region may be arbitrary description using a number, an alphabet, a symbol, etc., if it is unique.

  Next, the reference edge image generation process in step S105 will be described. FIG. 7 is a flowchart illustrating a processing procedure of reference edge image generation processing. As shown in FIG. 7, first, the smoothing unit 4a performs a smoothing process for smoothing the cell fluorescence image acquired from the image acquisition unit 2 (step S131). And the structure preservation | save smoothing part 4b performs the structure preservation | save smoothing process which further reduces the noise inside a cell area | region with respect to this smoothed cell fluorescence image (step S133).

  Subsequently, the edge extraction unit 4c performs edge extraction processing for detecting an edge from the cell fluorescence image smoothed by the structure preserving smoothing unit 4b and generating an edge image (step S135). The contour emphasizing unit 4d performs contour emphasis processing for emphasizing an edge component indicating the contour portion of the cell region among the edges in the edge image (step S137). The gradation converting unit 4e performs gradation conversion processing for generating a reference edge image by performing gradation conversion based on the edge image after the contour enhancement processing (step S139). Then, the control unit 8 causes the image conversion unit 5 to output the reference edge image from the gradation conversion unit 4e, and then proceeds to the process of step S107.

  In step S131, the smoothing unit 4a reduces fine luminance unevenness in the cell fluorescence image by, for example, convolving a Gaussian filter shown in FIG. 8 with the cell fluorescence image. Moreover, in step S133, the structure preservation | save smoothing part 4b performs the process which converts the luminance value of a process target pixel with reference to the 5 * 5 pixel centering on a process target pixel, for example, for each pixel in a cell fluorescence image. The entire cell fluorescence image is smoothed by sequentially performing processing as pixels to be processed.

ここで、変数ａ(ｉ)（ｉ＝１〜２５）は、変数ｓｖ(ｉ)を用いた次式（２），（３）によって求められる。なお、式（２）では、演算値ｍａｘ（ｓｖ）≠０であることが前提とされている。また、式（２）で用いる指数αは、α＝０．５とすることが好ましい。

ｓｖ(ｉ) ＝ ｜ｖ(ｉ)−ｖ(１３)｜ ・・・（３） FIG. 9 is a schematic diagram showing 5 × 5 pixels referred to by the structure preserving smoothing unit 4b in the smoothing process. In the figure, a variable indicating the luminance value of each pixel is shown in a rectangle indicating each pixel. The structure preserving smoothing unit 4b converts the luminance value v (13) of the processing target pixel at the center of the 5 × 5 pixels into the luminance value v ′ (13) by the following equation (1). By this conversion processing, a weighted average of luminance values is obtained by increasing the weight of neighboring pixels whose luminance values are closer to the processing target pixel.

Here, the variable a (i) (i = 1 to 25) is obtained by the following equations (2) and (3) using the variable sv (i). In formula (2), it is assumed that the calculated value max (sv) ≠ 0. The index α used in the formula (2) is preferably α = 0.5.

sv (i) = | v (i) −v (13) | (3)

  In addition, the smoothing process of step S131 and the structure preservation | save smoothing process of step S133 can be abbreviate | omitted according to the kind etc. of the cell fluorescence image acquired from the image acquisition part 2. FIG.

In step S135, the edge extraction unit 4c calculates a primary differential value and a secondary differential value for each pixel in the cell fluorescence image smoothed by the structure preserving smoothing unit 4b. When using the primary differential value, the edge extraction unit 4c calculates using, for example, a Sobel filter that is a primary differential filter. In this case, the edge extraction unit 4c uses the convolution output S1 (k) of the x-direction Sobel filter shown in FIG. 10-1 and the convolution output S2 (k) of the y-direction Sobel filter shown in FIG. 10-2. Then, the primary differential value E ₁ (k) of the pixel number k is calculated by the following equation (4). Here, the pixel number k indicates the pixel number of each pixel.

Moreover, when using a secondary differential value, the edge extraction part 4c calculates using the secondary differential filter which detects the recessed part of a luminance value change, for example. In this case, the edge extraction unit 4c outputs the convolution output C1 (k) of the x-direction concave portion detection filter shown in FIG. 11-1 and the convolution output C2 (k) of the y-direction concave portion detection filter shown in FIG. 11-3, the convolution output C3 (k) of the x + y + direction concave portion detection filter shown in FIG. 11 and the convolution output C4 (k) of the x + y− direction concave portion detection filter shown in FIG. The maximum value is extracted from the inside as shown in the following equation (5), and is set as the secondary differential value E ₂ (k) of the pixel number k. Here, the pixel number k indicates the pixel number of each pixel.
E ₂ (k) = Max (C1 (k), C2 (k), C3 (k), C4 (k)) (5)

In step S137, the contour emphasizing unit 4d determines the luminance value E (k) of each pixel of the edge image generated by the edge extracting unit 4c and the luminance of each pixel of the cell fluorescence image smoothed by the structure preserving smoothing unit 4c. Using the value V (k), the luminance value E _c (k) of each pixel in which the edge component of the contour portion of the cell region is emphasized is calculated by the following equation (6). Note that the coefficient m used in Equation (6) is preferably m = 0.55.
E _c (k) = m · E (k) −V (k) (6)

In step S139, as shown in FIG. 12A, the gradation conversion unit 4e applies four types of directional filters FT1 to FT4 to the central processing target pixel, and the luminance values of the pixels in each directional filter. The maximum luminance value Ec _max and the minimum luminance value Ec _min are extracted from the above. Then, gradation conversion is performed so that the extracted maximum luminance value Ec _max is 1 and the minimum luminance value Ec _min is 0. At this time, the gradation conversion unit 4e converts the luminance value Ec _{in of the} pixel to be processed into the luminance value Ec _out using a gradation conversion function expressed by the following equation (7). FIG. 12-2 shows the relationship between the luminance value Ec _in as an input to the tone conversion function and the luminance value Ec _out as an output. The index γ used in the equation (7) is preferably γ = 4.

  The direction filter FT1 is a horizontal filter that refers to the horizontal direction of the processing target pixel, and the direction filter FT2 is a vertical filter that refers to the vertical direction of the processing target pixel. The direction filter FT3 is an oblique direction filter that refers to the lower right and upper left directions, and the direction filter FT4 is an oblique direction filter that refers to the upper right and lower left directions. Each direction filter FT1 to FT4 refers to 10 pixels in each direction centered on the processing target pixel. However, the number of pixels to be referred to is not limited to 10 pixels in each direction, and can be increased or decreased. Each direction filter FT1 to FT4 is shown in FIG. 12A so as to refer to only one pixel column in each direction. However, the pixel column to be referred to is not limited to one column, and is subject to processing. It is also possible to refer to three or more pixel columns centering on the pixel column including the pixels.

  An example of the reference edge image generated by the reference edge image generation process of step S105 described above is shown in FIG. Further, FIG. 13-2 shows a cell fluorescence image corresponding to the reference edge image shown in FIG.

  Next, the recognition reliability determination process in step S107 will be described. In this recognition reliability determination process, the recognition reliability determination unit 5d determines the recognition result for each cell region based on the contour edge component intensity average and the internal edge component intensity average of each cell region in the recognition label image. The degree of success or failure of recognition of each cell region is determined by determining the degree of goodness and the component that inhibits recognition.

  The degree of success or failure of recognition of the recognized cell region can be said to be higher as the edge component of the contour portion is larger and the edge component inside the cell region is smaller. That is, if the edge component of the contour portion is small, the cell region is considered to have a high possibility of being formed by dividing a part of a larger cell region. In addition, if the edge component inside the cell region is large, the cell region is highly likely to be divided into finer cell regions, or is likely to be a cell region containing a lot of noise. Therefore, the recognition reliability determination unit 5d determines the recognition reliability, which is the validity of the recognized cell region and indicates the degree of success or failure of the recognition by the cell recognition processing, as the edge component of each cell region and the inside of the cell region. Judgment is made based on the size of the edge component.

FIG. 14 is a flowchart illustrating a processing procedure of recognition reliability determination processing. As shown in FIG. 14, first, the recognition label image separation unit 5 a separates the cell region in the recognition label image acquired from the cell recognition unit 3 into the contour portion and the inside of the region, and the contour portion label image and the internal label image. Are generated (step S141). Then, the contour edge component calculation unit 5b calculates the contour edge component strength average V _con (L) based on the contour label image (step S143), and the internal edge component calculation unit 5c calculates the internal label image. The inner edge component intensity average V _in (L) is calculated based on the original (step S145). In the first embodiment, the degree of success or failure of recognition of each cell region is determined based on the contour edge component intensity average V _con (L) and the internal edge component intensity average V _in (L) described in detail later. .

Subsequently, the recognition reliability determination unit 5d determines whether the contour edge component strength average V _con (L) is smaller than a predetermined threshold ContourTh and the inner edge component strength average V _in (L) is larger than a predetermined threshold InnerTh. It is determined whether or not the recognition of each cell region is successful (step S147). When this determination condition is satisfied (step S147: Yes), the recognition reliability determination unit 5d assigns “recognition reliability 4” to the cell region whose label number is the number L as the region marker (step S149).

When the determination condition of step S147 is not satisfied (step S147: No), the recognition reliability determination unit 5d determines whether the inner edge component strength average V _in (L) is larger than the threshold InnerTh, and the recognition success / failure is determined. Is further determined (step S151). When this determination condition is satisfied (step S151: Yes), “recognition reliability 3” is assigned to the cell region with the label number L (step S153).

When the determination condition of step S151 is not satisfied (step S151: No), the recognition reliability determination unit 5d determines whether the contour edge component strength average V _con (L) is smaller than the threshold value ContourTh, and performs recognition. The degree of success or failure is further determined (step S155). When this determination condition is satisfied (step S155: Yes), “recognition reliability 2” is assigned to the cell region with the label number L (step S157). If the determination condition of step S155 is not satisfied (step S155: No), “recognition reliability 1” is assigned to the cell region of label number L (step S159).

  Then, the control unit 8 determines whether or not all cell regions in the recognition label image have been processed, that is, whether or not recognition reliability is assigned to all the cell regions (step S161). If not (step S161: No), the process from step S141 is repeated for the unallocated cell region. On the other hand, when the recognition reliability is assigned to all the cell regions (step S161: Yes), the control unit 8 displays the recognition label image associated with the recognition reliability from the recognition reliability determination unit 5d. After output to the unit 6, the process proceeds to step 109. In this recognition reliability determination processing procedure, the control unit 8 can also execute steps S143 and S145 by switching the processing order.

  In step S141, as shown in FIG. 15, the recognition label image separation unit 5a includes a contour region that includes pixels that belong to the cell region and that are in contact with the background or other cell regions. The cell region is separated into the region composed of internal pixels excluding the contour portion. However, the recognition label image separation unit 5a repeats the operation of removing the contour portion n times so that a gap corresponding to the number of pixels n is provided between the contour portion and the inside of the region. The number of pixels n is preferably about n = 3. The gap between the contour portion and the inside of the region is a measure for reliably removing pixels corresponding to the contour portion from the inside of the region.

ラベル番号Ｌの細胞領域の輪郭部を構成する画素数ｎ_con(Ｌ)は、次式（９）に示すように、輪郭部ラベル画像におけるラベル番号Ｌの画素数をカウントすることによって求められる。

そして、輪郭部エッジ成分演算部５ｂは、式（８）および式（９）の結果を用いて、次式（１０）によって、輪郭部エッジ成分強度平均Ｖ_con(Ｌ)を算出する。

In step S143, the contour edge component calculation unit 5b first applies the edge pixel value of the pixel in the reference edge image corresponding to each contour pixel in the contour label image to the cell region with the label number L. Using the contour edge strength Ec _out (p), an edge component total SV _con (L) indicating the sum of the edge strengths is calculated by the following equation (8). Here, the variable p indicates the pixel number of the contour pixel.

The number of pixels n _con (L) constituting the contour portion of the cell region with the label number L can be obtained by counting the number of pixels with the label number L in the contour portion label image as shown in the following equation (9).

Then, the contour edge component calculation unit 5b calculates the contour edge component intensity average V _con (L) by the following equation (10) using the results of the equations (8) and (9).

ラベル番号Ｌの細胞領域内部を構成する画素数ｎ_in(Ｌ)は、次式（１２）に示すように、内部ラベル画像におけるラベル番号Ｌの画素数をカウントすることによって求められる。

そして、内部エッジ成分演算部５ｃは、式（１１）および式（１２）の結果を用いて、次式（１３）によって、内部エッジ成分強度平均Ｖ_in(Ｌ)を算出する。

In step S145, as in step S143, the internal edge component calculation unit 5c first applies to the cell region with the label number L, the edge of the pixel in the reference edge image corresponding to each internal pixel in the internal label image. Using the in-region edge strength Ec _in (q) as the pixel value, an edge component total SV _in (L) indicating the sum of the edge strengths is calculated by the following equation (11). Here, the variable q indicates the pixel number of the internal pixel.

The number of pixels n _in (L) constituting the inside of the cell region with the label number L can be obtained by counting the number of pixels with the label number L in the internal label image as shown in the following equation (12).

Then, the inner edge component calculation unit 5c calculates the inner edge component intensity average V _in (L) by the following equation (13) using the results of the equations (11) and (12).

  Next, the display of the recognition result in step S109 will be described. FIG. 16 is an image showing a recognition result based on the reference edge image and the cell fluorescence image shown in FIGS. 13A and 13B, and shows an example of the image displayed on the display unit 6 in step S109. is there. As illustrated in FIG. 16, the control unit 8 performs control to display each cell region in the recognition label image so as to be identifiable in a manner corresponding to the recognition reliability associated with the recognition reliability determination unit 5 d.

  Specifically, in FIG. 16, the cell region associated with “recognition reliability 1” indicating that the recognition success level is the highest is drawn in black inside the region (for example, cell region AR1). . The cell regions associated with “recognition reliability 2” and “recognition reliability 3” are drawn in dark gray (for example, cell region AR2) and vertical stripe pattern (for example, cell region AR3), respectively. The cell region associated with “recognition reliability 4” indicating that the recognition success level is the lowest is drawn in light gray (for example, cell region AR4).

  Note that the recognition result display method need not be interpreted as limited to the example of FIG. 16, and a different number, symbol, etc. is associated with each cell area for each recognition reliability, and is converted into a color image for recognition reliability. As long as the cell region can be identified for each recognition reliability, such as associating different colors with each other or associating different patterns, any display method can be applied.

  By displaying each cell region in association with the recognition reliability in this way, the observer can easily discriminate a cell region that has a high possibility of causing erroneous recognition. When there are many cell regions with low recognition reliability over the entire recognized region, there is a problem in the setting of processing parameters at the time of cell recognition processing, or there is a lot of noise in the cell fluorescence image, etc. The control unit 8 notifies the observer of warning information to that effect. Furthermore, the control part 8 can also alert | report warning information corresponding to each cell area | region with low recognition reliability.

  As described above, in the image processing device 1 and the image processing program according to the first embodiment, the cell region corresponding to each cell is automatically recognized from the cell fluorescence image, and the recognized cell region is recognized. Since the degree of success or failure is determined, it is possible to quickly determine how well each cell region is recognized by the cell recognition process, and it is easy to pick up the problem area, making it more reliable. High recognition results can be obtained. In addition, as a result, it is possible to perform processing according to the degree of success or failure of recognition of each cell region in various processing performed based on the recognition result, such as measurement processing of the feature amount of the recognized cell region, cell tracking processing, etc. To do.

  Further, generally, when an observer visually observes a cell image, it is considered that a portion perceived to have changed in luminance with respect to the surroundings is captured as the outline of the cell, and the first embodiment is applied. In the image processing apparatus 1 and the image processing program, since the recognition reliability is associated using the reference edge image obtained by extracting the perceived portion as an edge component, the recognition reliability matched with the observer's perception is associated. It is possible.

(Embodiment 2)
Next, a second embodiment of the present invention will be described. In the second embodiment, after recognizing the cells and associating the recognition reliability, the cells are traced between the cell fluorescence images at different observation points.

  FIG. 17 is a block diagram showing a configuration of the image processing apparatus 11 according to the second embodiment of the present invention. As shown in FIG. 17, the image processing apparatus 11 includes a control unit 18 instead of the control unit 8 and a cell tracking unit 19 based on the configuration of the image processing apparatus 1. Each component including the cell tracking unit 19 is electrically connected to the control unit 18. In addition, the same code | symbol is attached | subjected to the component same as Embodiment 1. FIG.

  In the image processing device 11, the image acquisition unit 2 acquires a series of cell fluorescence images at different observation points from an external device, and each of the cell recognition unit 3, the reference edge image generation unit 4, and the image matching unit 5 For each cell fluorescence image, a recognition label image, a reference edge image, and a recognition reliability are associated. FIG. 18 shows an example of a series of cell fluorescence images, a reference edge image, and a recognition label image drawn according to the recognition reliability.

  The cell tracking unit 19 acquires the recognition label image corresponding to the series of cell fluorescence images, the parameters used for determination of the recognition reliability and the recognition reliability from the image matching unit 5, and determines the degree of success or failure of the recognition of each cell region. Reflecting this, cell identity between cell images at different observation points is determined, and cell tracking processing for associating cell regions including cells determined to be the same is performed. In addition, the cell tracking unit 19 outputs a tracking result as a processing result of the cell tracking process to the display unit 6. Note that the cell tracking unit 19 can also store the tracking result in the storage unit 7 via the control unit 18.

  The control unit 18 controls the cell tracking unit 19 in addition to controlling the processing and operation of each unit in the same manner as the control unit 8. In particular, the control unit 18 executes the image processing program stored in the storage unit 7 and further performs a cell tracking process between step S107 and step S109 based on the image processing procedure shown in FIG. A series of image processing is controlled. In addition, the control unit 18 performs control to display the tracking result output from the cell tracking unit 19 on the display unit 6 as image information or numerical information in association with the recognition label image, the recognition reliability, and the like at each observation time point. Do.

  Here, the process performed by the cell tracking unit 19 will be described in detail. In the cell tracking process, the cell tracking unit 19 calculates the correspondence between the cell regions between two recognition label images at different observation points by reflecting the degree of success or failure of the recognition of each cell region, and the calculated correspondence Based on the above, the identity of the cells in each cell region is determined. At this time, the cell tracking unit 19 calculates the correspondence using, for example, the overlapping area of the overlapping portions of the cell regions.

More specifically, cell tracking unit 19, and cell region recognition label image of the observation time t ₁ is a label number Ri _t1, the continuous time series and observation time t ₁ observation time t ₂ recognition label image and cell area is a label number Ri _t2 in the inner of the corresponding degree M of (Ri _t1, Ri _t2), the overlapping area OL_Area of this cell between regions (Ri _t1, Ri _t2), the area area (Ri for each cell region _t1), by using the Area (Ri _t2), calculated by the following equation (14). Note that the coefficient w in Expression (14) is a predetermined weighting coefficient.

In addition, the cell tracking unit 19 is not limited to not only between time-series continuous recognition label images but also a predetermined number of frames away from each other in order to reduce the influence of temporal changes in the amount of fluorescence that occurs when capturing a series of cell fluorescence images. The degree of correspondence is also calculated between the recognition label images. Labels In this case, cell tracking unit 19, and a cell region is the label number Ri _ta the recognition label image corresponding to the observation time t _a, in the recognition label image of a predetermined number b only the previous frame from the recognition label image and cell area is number Ri _{t (ab),} the corresponding degree _{M (Ri t (ab),} Ri ta) the overlapping area OL_Area of this cell between regions _{(Ri t (ab), Ri} ta) and, each cell Using the area of the area Area (Rit _(ab) ) and Area ( _Rita ), it is calculated by the following equation (15). Note that the coefficient w _f in the equation (15) is a predetermined weighting coefficient.

Furthermore, the cell tracking unit 19 calculates a correspondence degree reflecting the degree of success or failure of recognition of each cell region based on the correspondence degree calculated by Expression (15). For this reason, the second embodiment reflects the internal edge component intensity average V _in (L) and the contour edge component intensity average V _con (L), which are criteria for determining the degree of success or failure of recognition of each cell region. The degree of correspondence is calculated. That is, cell tracking unit 19, the label number Ri _ta corresponding degree M of the degree of success or failure of recognition of a cell region and the label number Ri t cell area is _(ab) considering a '(Ri t _(ab), Ri the _ta), based on the corresponding degree M _(Ri t _(ab), and Ri _ta), label number Ri t _(ab) a is the internal edge component intensity average V _in the cell region _(Ri t _(ab)) and contours Using the partial edge component intensity average V _con (Rit _(ab) ), the following equation (16) is used.

Cell tracking unit 19, such corresponding degree _{M '(Ri t (ab)} , Ri ta) and a cell region existing recognition label image of the observation time t _a, the observation time t _(ac) (however, c = 1 The calculation is performed for all possible combinations of the cell region existing in the recognition label image at each observation time in (b). Then, it is determined that each cell in the combination of the cell regions showing the highest degree of correspondence is the same cell, and the cell regions in this combination are associated with each other. However, if all the corresponding degree calculated between the observed time t _a and the observation time t _(ac) is less than 0 or a predetermined threshold, cell tracking unit 19, corresponds between these observations point attached cells do not exist, the cells at the observation point t _a is determined that the newly generated.

  Here, the cell tracking unit 19 calculates the correspondence using the overlapping area of the cell regions. However, the cell tracking unit 19 is not limited to the overlapping area. It can also be calculated using other feature quantities.

  As described above, in the image processing apparatus 11 and the image processing program according to the second embodiment, the identity of cells between cell fluorescence images at different observation points is reflected by reflecting the degree of success or failure of recognition of each cell region. Therefore, it is possible to prevent propagation of an error generated in the cell recognition process and realize a more reliable cell tracking process.

(Embodiment 3)
Next, a third embodiment of the present invention will be described. In the first embodiment described above, the degree of success or failure of the recognition is determined only in the cell region as a result of the integration of the temporary cell region. However, in this third embodiment, the temporary cell region is temporarily determined according to a predetermined condition. A cell region is selected to determine the degree of success or failure of recognition.

  FIG. 19 is a block diagram showing a configuration of the image processing apparatus 21 according to the third embodiment of the present invention. As shown in FIG. 19, the image processing device 21 is replaced with the cell recognition unit 3, the image matching unit 5, and the control unit 8 based on the configuration of the image processing device 1, and the cell recognition unit 23 and the image matching unit 25. And a control unit 28, and an integrated outline extraction unit 29 and a selective integrated label image generation unit 30 are newly provided. Each component including the integrated contour extraction unit 29 and the selective integrated label image generation unit 30 is electrically connected to the control unit 28. In addition, the same code | symbol is attached | subjected to the component same as Embodiment 1. FIG.

  The cell recognizing unit 23 is based on the configuration of the cell recognizing unit 3 provided in the image processing apparatus 1, and the background removing units 23e-1 and 23e- are obtained by dividing the background removing unit 3e and the labeling unit 3f into two components. 2 and label portions 23f-1 and 23f-2. Then, the region integration unit 3d, the background removal unit 23e-1 and the labeling unit 23f-1 perform steps similar to the cell recognition process shown in FIG. 4 for the temporary label image generated by the ultimate value calculation unit 3c. Steps S117 to S121 are performed to generate a recognition label image. On the other hand, the background removal unit 23e-2 and the labeling unit 23f-2 omit step S117 and perform steps S119 and S121 on the temporary label image to generate a temporary label image from which the background pixels are removed.

  The labeling units 23f-1 and 23f-2 respectively assign unique area markers to the recognition label image and the temporary label image. At this time, in a case where the labeling unit 23f-1 gives a region marker different from any of the region markers given to each temporary cell region by the ultimate value calculating unit 3c, the labeling unit 23f-2 The area label given by the calculation unit 3c can be used as it is. In this case, the labeling unit 23f-2 can be removed from the cell recognition unit 23.

  The integrated contour extraction unit 29 acquires the recognition label image and the temporary label image from the labeling units 23f-1 and 23f-2, and calculates the difference between the recognition label image and the temporary label image. As a result, the integrated contour extraction unit 29 extracts the integrated contour corresponding to the boundary line between the temporary cell regions integrated by the region integration unit 3d, for example, as shown in FIG. 20, and shows the extracted integrated contour An integrated outline image is generated.

  The selective integrated label image generation unit 30 integrates the recognition label image and the temporary label image from the label units 23f-1 and 23f-2, the reference edge image from the reference edge image generation unit 4, and the integrated outline extraction unit 29 to be integrated. Each contour image is acquired. Then, according to the edge pixel value of the pixel in the reference edge image corresponding to the integrated contour pixel as the boundary pixel included in the integrated contour in the integrated contour image, the cell region or the temporary label in the recognition label image The temporary cell region in the label image is recognized as the final cell region.

  More specifically, the selective integrated label image generation unit 30, when the edge pixel value of the pixel in the reference edge image corresponding to the integrated contour pixel constituting the integrated contour is larger than a predetermined threshold, Each adjacent temporary cell region is recognized as a final cell region with this integrated contour in the temporary label image as a boundary line. This is because when the edge strength corresponding to the integrated contour is large, the integrated contour is highly likely to indicate the contour of the cell, and each temporary cell region is likely to be a region corresponding to an individual cell. It depends on what is considered.

  In addition, the selective integrated label image generation unit 30 integrates adjacent temporary cell regions when the edge pixel value of the pixel in the reference edge image corresponding to the integrated contour pixel is equal to or less than a predetermined threshold. Recognize the cell region as the final cell region. This is because when the edge strength corresponding to the integrated contour is small, the integrated contour is likely to show a luminance change other than the cell contour, and each adjacent temporary cell region corresponds to the same cell. It is because it is considered that there is a high possibility.

  Further, the selective integrated label image generation unit 30 generates a selective integrated label image indicating the final cell region recognized as described above. At this time, whether each cell region in the selectively integrated label image is a temporary cell region or a cell region as a result of integration is recorded for each cell region. In addition, the selective integrated label image generation unit 30 outputs the generated selective integrated label image to the image matching unit 25. The selective integrated label image generation unit 30 can also output and store the generated selective integrated label image to the storage unit 7 via the control unit 28.

The image matching unit 25 uses the selective integrated label image instead of the recognition label image, performs the recognition reliability determination process in the same manner as the image matching unit 5 provided in the image processing apparatus 1, and performs the process in the selective integrated label image. The degree of success or failure of recognition is determined for each cell region. However, when performing the calculation shown in the equations (11) to (13) for the cell region as the integration result, the image matching unit 25 uses the integrated contour instead of using all the pixels constituting the inside of the cell region. The calculation is performed using only the _in- region edge intensity Ec _in (q) corresponding to the pixel, and the number of pixels n _in (L) is calculated as the number of integrated contour pixels. By performing the arithmetic processing in this way, it is possible to realize the efficiency and speedup of the recognition reliability determination processing.

  Here, a processing procedure performed by the image processing apparatus 21 will be described. FIG. 21 is a flowchart illustrating an image processing procedure in which the image processing device 21 processes a cell fluorescence image by the control unit 28 executing the image processing program. As shown in FIG. 21, first, the image acquisition unit 2 acquires a cell fluorescence image from the external device (step S201), and the cell recognition unit 23 performs the temporary label image and the step similar to step S103 shown in FIG. 3. A cell recognition process for generating a recognition label image is performed (step S203).

  However, while only the recognized label image is output in step S103, in step S203, both the recognized label image and the temporary label image are output. Further, in steps S119 and S121 shown in FIG. 4 as internal processing in step S103, processing is performed only on the recognized label image. In step S203, the recognition label image and the temporary label image are processed. For both, the same processing as in step S119 and step S121 is performed.

  Subsequently, the reference edge image generation unit 4 performs Step S205 in the same manner as Step S105 illustrated in FIG. The integrated contour extraction unit 29 calculates the difference between the recognized label image and the temporary label image acquired from the labeling units 23f-1 and 23f-2, extracts the integrated contour, and generates an integrated contour image ( Step S207). The selective integrated label image generation unit 30 selects a cell region in the recognition label image or a temporary cell region in the temporary label image according to the edge pixel value of the pixel in the reference edge image corresponding to the integrated contour pixel. Then, a selective integrated label image generation process for generating a selective integrated label image indicating the selected region as the final cell region is performed (step S209).

  Thereafter, the image matching unit 25 and the display unit 6 sequentially perform steps S211 and S213 similarly to steps S107 and S109 illustrated in FIG. 3, and the control unit 28 ends the series of image processing. However, in step S211, the image matching unit 25 performs processing using a selective integrated label image instead of the recognized label image.

  In step S213, the display unit 6 can newly display a selective integrated label image, an integrated outline image, and the like in step S109. In step S213, whether the display unit 6 has selected a temporary cell region or a cell region as a result of integration for each cell region in the selectively integrated label image under the control of the control unit 28. Can be notified to the observer as text information or image information. The control unit 28 may perform step S203 and step S205 by switching the processing order.

  As described above, in the image processing device 21 and the image processing program according to the third embodiment, when the cell region is recognized, the edge pixel value in the reference edge image of the integrated contour pixel, that is, the integrated contour is used. Depending on the corresponding edge strength, the cell area in the recognition label image or the temporary cell area in the temporary label image is selected to be the final cell area. Can be realized. As a result, it is possible to perform processing with higher reliability in various types of processing performed based on the recognition result, such as processing for measuring the feature amount of the recognized cell region and processing for tracking the cell.

  Further, in the image processing device 21 and the image processing program according to the third embodiment, when determining the degree of success or failure of recognition of each cell region, all of the cell regions inside the cell region as a result of integration are configured. Instead of using these pixels, various calculations are performed using only the integrated contour pixels, so that the recognition reliability determination process can be made more efficient and faster.

  Further, in the image processing device 21 and the image processing program according to the third embodiment, the size of the edge intensity corresponding to the integrated contour is determined based on a predetermined threshold value, so that the cell region of the integration result It is possible to evaluate the validity of the integration processing for the noise, the amount of noise in the cell region of the integration result, and the like.

(Embodiment 4)
Next, a fourth embodiment of the present invention will be described. In the first embodiment described above, the recognition label image and the reference edge image are collated to determine the recognition reliability. However, in the fourth embodiment, a predetermined region set for the cell region is further determined. The degree of success or failure of recognition of each cell region is determined according to the degree of matching with the model.

  FIG. 22 is a block diagram showing a configuration of the image processing apparatus 31 according to the fourth embodiment of the present invention. As shown in FIG. 22, the image processing device 31 includes an image matching unit 35 and a control unit 38 instead of the image matching unit 5 and the control unit 8 based on the configuration of the image processing device 1. FIG. 23 is a block diagram illustrating a detailed configuration of the image matching unit 35. As shown in FIG. 23, the image matching unit 35 includes a recognition reliability determination unit 35d instead of the recognition reliability determination unit 5d based on the configuration of the image verification unit 5, and includes a feature amount calculation unit 35e, a cell A model assigning unit 35f and a model suitability calculating unit 35g are newly provided. In addition, the same code | symbol is attached | subjected to the component same as Embodiment 1. FIG.

  Similar to the image matching unit 5, the image matching unit 35 calculates the contour edge component strength average and the inner edge component strength average based on the recognition label image and the reference edge image to recognize each cell region. The degree of success or failure is determined. In addition to this processing, the image matching unit 35 determines the degree of success or failure of recognition of each cell region according to the degree of matching between the cell region in the recognition label image and the region model set in advance for the cell region. Make a decision. Here, the region model set in advance is defined based on the shape of the cell region, the luminance distribution, and the like.

  More specifically, the cell model assigning unit 35f sets a predetermined area model for the cell area. The model suitability calculation unit 35g sets a threshold value as a criterion for the suitability according to the set area model. On the other hand, the feature amount calculation unit 35e calculates a feature amount necessary for determining the degree of fitness with the region model for each cell region in the recognition label image, and outputs the feature amount to the model suitability calculation unit 35g. The recognition reliability determination unit 35d compares the threshold value as the determination criterion acquired from the model suitability calculation unit 35g with the feature amount of the cell region. Assign recognition confidence. Further, when the degree of matching is low, similarly to the recognition reliability determination unit 5, a process of assigning recognition reliability based on the contour edge component strength average and the internal edge component strength average is performed.

  Here, as a process performed by the image processing apparatus 31, a recognition reliability determination process corresponding to step S107 illustrated in FIG. 3 will be described. The image processing procedure in which the image processing apparatus 31 processes the cell fluorescence image by executing the image processing program by the control unit 38 is the same as the image processing procedure shown in FIG. 3 except for step S107.

  FIG. 24 is a flowchart illustrating a processing procedure of recognition reliability determination processing performed by the image processing device 31. This flowchart exemplifies a recognition reliability determination process when a circular area model is set by the cell model assigning unit 35f and a threshold RoundTh for the circularity is set by the model suitability calculating unit 35g.

As shown in FIG. 24, first, the feature amount calculation unit 35e calculates the circularity Rd _L of the cell region with the label number L as the feature amount necessary for determining the degree of compatibility with the region model (step S221). ). Subsequently, the recognition reliability determination unit 35d determines whether or not the circularity Rd _L is greater than or equal to the threshold value RoundTh (step S223). When the circularity Rd _L is greater than or equal to the threshold RoundTh (step S223: Yes), the recognition reliability determination unit 35d assigns “recognition reliability 1” to the cell region with the label number L (step S225).

When the circularity Rd _L is smaller than the threshold RoundTh (step S223: No), the recognition label image separation unit 5a, the contour edge component calculation unit 5b, and the internal edge component calculation unit 5c are the same as steps S141 to S145 shown in FIG. Similarly, steps S227 to S231 are executed. Then, the recognition reliability determination unit 35d determines whether the contour edge component strength average V _con (L) is smaller than the predetermined threshold ContourTh and whether the inner edge component strength average V _in (L) is larger than the predetermined threshold InnerTh. (Step S233), and if this determination condition is satisfied (Step S233: Yes), “recognition reliability 5” is assigned to the cell region of the label number L (Step S235).

When the determination condition of step S233 is not satisfied (step S233: No), the recognition reliability determination unit 35d determines whether or not the inner edge component strength average V _in (L) is greater than the threshold InnerTh (step S237). When this determination condition is satisfied (step S237: Yes), “recognition reliability of 4” is assigned to the cell region with the label number L (step S239).

When the determination condition of step S237 is not satisfied (step S237: No), the recognition reliability determination unit 35d determines whether or not the contour edge component strength average V _con (L) is smaller than the threshold value ContourTh (step S241). If this determination condition is satisfied (step S241: Yes), “recognition reliability 3” is assigned to the cell region with the label number L (step S243). If the determination condition of step S241 is not satisfied (step S241: No), “recognition reliability 2” is assigned to the cell region of label number L (step S245).

  And the control part 38 judges whether all the cell areas in the recognition label image were processed (step S247), and when not processing all the cell areas (step S247: No), it is processing. The process from step S221 is repeated for a cell region that does not exist. On the other hand, when all the cell regions are processed (step S247: Yes), the control unit 38 outputs the recognition label image associated with the recognition reliability to the display unit 6 from the recognition reliability determination unit 35d. Then, the recognition reliability determination process is terminated. In this recognition reliability determination processing procedure, the control unit 38 can also execute step S229 and step S231 by switching the processing order.

In step S221, the feature amount calculation unit 35e calculates the circularity Rd _L by the following equation (17) using the area S of the cell region and the perimeter T.
Rd _L = 4πS / T (17)

In this recognition reliability determination process, the suitability of the cell region with respect to the region model is preferentially determined in steps S221 to S225, and if the circularity Rd _L is equal to or greater than the threshold RoundTh and close to a perfect circle, an edge is determined. “Recognition reliability 1” indicating that the recognition success level is the highest is associated with the high matching degree regardless of the strength.

  In addition, the recognition label image associated with the recognition reliability by the recognition reliability determination process is identified for each cell region according to the recognition reliability under the control of the control unit 38 as in the first embodiment. Displayed as possible.

Next, as another example of the recognition reliability determination process performed by the image processing device 31, a recognition reliability determination process when a region model having a convex luminance distribution is set by the cell model assignment unit 35f will be described. To do. In this case, for example, the model fitness calculating part 35g for the presence ratio r _c to the entire cell area of the convex region luminance distribution is convex in the cell region to set the threshold ConvexityTh, feature amount calculation unit 35e is It shall be calculated abundance ratio r _c with respect to each cell area in the recognition label image. Such setting of the region model and the threshold ConvexityTh is based on the fact that the cell region generally appears as a cluster of high-luminance pixels in the cell fluorescence image.

More specifically, the feature amount calculation unit 35e first detects a convex region from the cell region in the recognition label image. At this time, for example, as shown in FIG. 25, reference is made to pixels A and B that are symmetrically separated by a predetermined distance D with respect to the target pixel P. If the pixel value I _P of the pixel of interest _P is larger than the average luminance value I _{AB of the} luminance values I _A and I _B of the pixels A and B, the candidate that constitutes the convex region when the pixel of interest _P is large. It detects as a pixel. The feature quantity calculation unit 35e performs such detection processing in the four directions shown in FIG. 26, that is, the horizontal direction (I _Ahor -I _Bhor direction), the vertical direction (I _Aver -I _Bver direction), and the diagonally upward direction ( I _Asla -I _Bsla direction) and slanting left-down direction (I _Abac -I _Bbac direction), and when all the conditions expressed by the following equations (18) to (21) are satisfied, Are finally detected as the pixels constituting.
I _P > I _Mhor = (I _Ahor + I _Bhor ) / 2 (18)
I _P > I _Mver = (I _Aver + I _Bver ) / 2 (19)
I _P > I _Msla = (I _Asla + I _Bsla ) / 2 (20)
I _P > I _Mbac = (I _Abac + I _Bbac ) / 2 (21)

Subsequently, the feature amount calculation unit 35e determines whether or not all the pixels in the recognition label image are pixels constituting the convex region, and records the luminance convexity of the convex region. Generate an image. Here, the size C _P of the luminance projection in the pixel of interest P is calculated by the following equation (22).
C _P = ((I _P −I _Mhor ) + (I _P −I _Mver ) + (I _P −I _Msla ) + (I _P −I _Mbac )) / 4 (22)

After that, the feature amount calculation unit 35e determines whether or not the value of each pixel in the luminance convex image is larger than a predetermined threshold value, and binarizes the value as “1” when it is larger and “0” when it is not larger. A luminance convex binarized image is generated. Then, for the cell region with label number L in the recognition label image, the number n _L of pixels whose value in the luminance convex binarized image corresponding to the pixel constituting the cell region is “1” is calculated, and the calculated pixel The abundance ratio r _c (L) is calculated by the following equation (23) using the number n _L and the area S _{L of the} cell region.
r _c (L) = n _L / S _L (23)

As the recognition reliability determination processing procedure, the feature quantity calculation unit 35e calculates the abundance ratio r _c (L) of the cell region with the label number L instead of step S221 based on the flowchart shown in FIG. The recognition reliability determination unit 35d may change the processing content so as to determine whether or not the existence ratio r _c (L) is larger than the threshold value ConvexityTh instead of step S223. As a result, “recognition reliability 1” can be assigned to a cell region that is determined to have a high degree of fitness for a region model having a convex luminance distribution.

  As described above, in the image processing device 31 and the image processing program according to the fourth embodiment, the region model defined based on the shape of the cell region, the luminance distribution, etc. is set, and the fitness of the cell region with respect to the region model Since a high recognition reliability is associated with a cell region having a high degree of fitness, it is possible to efficiently discriminate a cell region with a high degree of recognition success that matches the region model. it can. In addition, a cell region having a high recognition success level that matches the region model can be displayed separately from other cell regions, and can be easily identified by an observer.

  In Embodiments 1 to 4 described above, the luminance value is particularly handled as the pixel value. However, the pixel value in the present invention is not limited to the luminance value. Intensity values etc. are included. The image processing apparatus according to the present invention can appropriately select a value as a pixel value from among these various values according to the property of the image to be processed.

1 is a block diagram illustrating a configuration of an image processing apparatus according to a first embodiment of the present invention. It is a block diagram which shows the detailed structure of the image collation part shown in FIG. 3 is a flowchart illustrating a processing procedure performed by the image processing apparatus illustrated in FIG. 1. It is a flowchart which shows the process sequence of the cell recognition process shown in FIG. It is a figure explaining the distance between extreme values. It is a figure explaining the integration method of a temporary cell area | region. FIG. 4 is a flowchart showing a processing procedure of reference edge image generation processing shown in FIG. 3. FIG. It is a figure which shows a Gaussian filter. It is a figure which shows the pixel which the structure preservation | save smoothing part shown in FIG. 1 refers. It is a figure which shows a primary differential filter. It is a figure which shows a primary differential filter. It is a figure which shows a secondary differential filter. It is a figure which shows a secondary differential filter. It is a figure which shows a secondary differential filter. It is a figure which shows a secondary differential filter. It is a figure which shows a direction filter. It is a figure which shows the input / output relationship of a gradation conversion function. It is a figure which shows an example of a reference edge image. It is a figure which shows an example of a cell fluorescence image. It is a flowchart which shows the process sequence of the recognition reliability determination process shown in FIG. It is a figure explaining the separation method of a cell field. It is a figure which shows an example of a recognition result. It is a block diagram which shows the structure of the image processing apparatus concerning Embodiment 2 of this invention. It is a figure which shows an example of the image processed with the image processing apparatus shown in FIG. It is a block diagram which shows the structure of the image processing apparatus concerning Embodiment 3 of this invention. It is a figure explaining the extraction method of a to-be-integrated outline. 20 is a flowchart illustrating a processing procedure performed by the image processing apparatus illustrated in FIG. 19. It is a block diagram which shows the structure of the image processing apparatus concerning Embodiment 4 of this invention. It is a block diagram which shows the detailed structure of the image collation part shown in FIG. It is a flowchart which shows the process sequence of the recognition reliability determination process which the image processing apparatus shown in FIG. 22 performs. It is a figure explaining the detection method of the area | region where luminance distribution is convex. It is a figure explaining the detection method of the area | region where luminance distribution is convex.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1,11,21,31 Image processing apparatus 2 Image acquisition part 3,23 Cell recognition part 3a Smoothing part 3b Gradient direction calculation part 3c Ultimate extreme value calculation part 3d Area | region integration part 3e, 23e-1, 23e-2 Background removal part 3f, 23f-1, 23f-2 Labeling unit 4 Reference edge image generating unit 4a Smoothing unit 4b Structure preserving smoothing unit 4c Edge extracting unit 4d Outline emphasizing unit 4e Tone conversion unit 5, 25, 35 Image matching unit 5a Recognition label image Separation unit 5b Contour part edge component calculation unit 5c Internal edge component calculation unit 5d, 35d Recognition reliability determination unit 6 Display unit 7 Storage unit 8, 18, 28, 38 Control unit 19 Cell tracking unit 29 Integrated contour extraction unit 30 Selection Integrated label image generation unit 35e feature quantity calculation unit 35f cell model assignment unit 35g model fitness calculation unit AR1 to AR4 cell region FT1 to FT4 Yilta

Claims (18)

In an image processing apparatus that performs recognition processing for recognizing a target area that is an image area corresponding to an observation target from among input observation images,
An image processing apparatus comprising: a recognition reliability determination unit that determines the degree of success or failure of recognition by the recognition process for each target region.   The recognition reliability determination unit determines the degree of success or failure of recognition by the recognition process using an edge pixel value of each pixel of an edge image in which an edge included in the observed image is detected. The image processing apparatus according to 1.   The recognition reliability determination means includes a contour edge strength that is an edge pixel value of a pixel in the edge image corresponding to a contour pixel included in a contour portion of the target region, and an inside of the region excluding the contour portion of the target region. The degree of success or failure of recognition by the recognition processing is determined using an in-region edge strength which is an edge pixel value of a pixel in the edge image corresponding to an internal pixel included in the recognition image. An image processing apparatus according to 1. Temporary area recognition means for recognizing a temporary target area from the observed image;
An area in which the adjacent temporary target areas are integrated to form an integrated area based on a feature amount indicating a feature between adjacent temporary target areas from among the temporary target areas recognized by the temporary area recognition unit. Integration means;
The temporary target area or the integrated area is set as the target area according to an edge pixel value of a pixel in the edge image corresponding to a boundary pixel included in a boundary line between the integrated temporary target areas in the integrated area. Selective area recognition means for recognizing;
The image processing apparatus according to claim 1, further comprising: Temporary area recognition means for recognizing a temporary target area from the observed image;
An area in which the adjacent temporary target areas are integrated to form an integrated area based on a feature amount indicating a feature between adjacent temporary target areas from among the temporary target areas recognized by the temporary area recognition unit. Integration means;
The temporary target area or the integrated area is set as the target area according to an edge pixel value of a pixel in the edge image corresponding to a boundary pixel included in a boundary line between the integrated temporary target areas in the integrated area. Selective area recognition means for recognizing;
Further comprising
The recognition reliability determination means determines the degree of success or failure of recognition by the recognition process using an edge pixel value of a pixel in the edge image corresponding to the boundary pixel as an edge strength in the region. The image processing apparatus according to claim 3.   The recognition reliability determination unit determines the degree of success or failure of recognition by the recognition process using a degree of matching between a target area recognized by the recognition process and a preset area model. Item 4. The image processing device according to any one of Items 1 to 3.   The image processing apparatus according to claim 6, wherein the region model is defined based on a shape or a luminance distribution of the observation target.   The display control means for performing control to display each target area recognized by the recognition processing in a mode according to the determination result of the recognition reliability determination means. The image processing apparatus according to one. The observation image includes a plurality of images taken at different observation time points,
It further comprises a target tracking unit that determines the identity of the observation target between observation images at different observation points, and the target tracking unit determines the degree of success or failure of recognition by the recognition process determined by the recognition reliability determination unit. The image processing apparatus according to claim 1, wherein the image processing apparatus is reflected in determination of identity of the observation target.   Corresponding to the target region determined to have a low recognition success level by the recognition reliability determination unit, the control unit further includes a notification control unit that performs control to notify warning information indicating that the recognition success level is low. The image processing apparatus according to claim 1, wherein the image processing apparatus is characterized in that:   The image processing apparatus according to claim 1, wherein the observation target is a cell. In an image processing program for causing an image processing apparatus that performs recognition processing to recognize a target region that is an image region corresponding to an observation target from among input observation images,
In the image processing apparatus,
An image processing program for executing a recognition reliability determination procedure for determining the degree of success or failure of recognition by the recognition processing for each target area.   The recognition reliability determination procedure determines a success or failure of recognition by the recognition process using an edge pixel value of each pixel of an edge image in which an edge included in the observed image is detected. 12. An image processing program according to 12.   The recognition reliability determination procedure includes: a contour edge strength which is an edge pixel value of a pixel in the edge image corresponding to a contour pixel included in a contour portion of the target region; and a region inside excluding the contour portion of the target region The degree of success or failure of recognition by the recognition processing is determined using an in-region edge strength that is an edge pixel value of a pixel in the edge image corresponding to an internal pixel included in the recognition image. The image processing program described in 1. A temporary area recognition procedure for recognizing a temporary target area from the observed image;
An area in which the adjacent temporary target areas are integrated to form an integrated area based on a feature amount indicating a feature between adjacent temporary target areas from among the temporary target areas recognized by the temporary area recognition procedure. Integration procedures and
The temporary target area or the integrated area is set as the target area according to an edge pixel value of a pixel in the edge image corresponding to a boundary pixel included in a boundary line between the integrated temporary target areas in the integrated area. Selective area recognition procedure to recognize;
The image processing program according to any one of claims 12 to 14, wherein the image processing program is further executed. A temporary area recognition procedure for recognizing a temporary target area from the observed image;
An area in which the adjacent temporary target areas are integrated to form an integrated area based on a feature amount indicating a feature between adjacent temporary target areas from among the temporary target areas recognized by the temporary area recognition procedure. Integration procedures and
The temporary target area or the integrated area is set as the target area according to an edge pixel value of a pixel in the edge image corresponding to a boundary pixel included in a boundary line between the integrated temporary target areas in the integrated area. Selective area recognition procedure to recognize;
Is executed further,
In the recognition reliability determination procedure, the degree of success or failure of recognition by the recognition process is determined using an edge pixel value of a pixel in the edge image corresponding to the boundary pixel as an edge strength in the region. The image processing program according to claim 14.   The recognition reliability determination procedure determines the success or failure of recognition by the recognition process using a degree of matching between a target area recognized by the recognition process and a preset area model. Item 15. The image processing program according to any one of Items 12 to 14. The observation image includes a plurality of images taken at different observation time points,
A target tracking procedure for determining the identity of the observation target between observation images at different observation points is further executed, and the target tracking procedure is a degree of success or failure of recognition by the recognition process determined in the recognition reliability determination procedure. The image processing program according to claim 12, wherein the image processing program is reflected in the determination of the identity of the observation target.

Images