JPH11167633A - Method and device for analyzing image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To properly perform image analysis based on a histogram when the exposure state of an image is 'back light'. SOLUTION: When the exposure state of an image of an original is discriminated and the discriminated exposure state is 'back light' (step S120), a mouse operation of an operator is received and the object that becomes a subject in the image is designated (step S200) and also, each peak in a doubly peaked histogram is divided into two (step S300). A division area side that includes a peaked part which represents the designated subject object is extracted and a histogram distribution in the extracted division area is made analyzed data (step S400). The analyzed data is used for image analytical processing which produces a LUT for density gradation return.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an image analysis method and apparatus for analyzing a histogram of image data representing an image as data to be analyzed.

[0002]

2. Description of the Related Art Conventionally, various image processes have been performed on image data using a computer for the purpose of improving the image quality of the image data. Various types of image processing are performed using appropriate processing parameters, and these processing parameters need to be obtained by analyzing image data before performing various types of image processing. In particular, when determining a reproduction range of an image as a processing parameter, it is common to perform analysis based on a histogram of image data.

In the analysis process using the histogram,
For example, a highlight point having the highest brightness in image data is selected, and a reproduction range is determined based on the highlight point.

[0004]

However, according to the above-mentioned prior art, when the exposure state of an image is a backlight, a portion of high brightness generated by the exposure state of the backlight, that is, the subject of the image (appropriate exposure) There is a problem in that erroneous analysis is performed, such as selecting the highlight point from an invalid portion distant from an image area desired to be reproduced in a state.
As a result, the reproduction range of the image data becomes inappropriate.

According to the present invention, the image exposure state is "backlit",
It is an object of the present invention to appropriately perform image analysis based on a histogram even in an inappropriate state such as "under exposure".

[0006]

Means for Solving the Problems and Actions and Effects Thereof, the following structure is adopted as means for solving at least a part of the above-mentioned problems.

According to the image analysis method of the present invention, there is provided an image analysis method in which a histogram of image data representing an image is used as data to be analyzed, and the image data is analyzed. A step of preparing a histogram having the above peaks; (b) a step of dividing the histogram prepared in the step (a) into a plurality of regions for each of the peaks;
The method includes the steps of: selecting one area from the plurality of divided areas; and (d) extracting a histogram distribution included in the selected area as the analyzed data. And

According to the image analysis method having this configuration, the histogram is divided into a plurality of regions for each peak, and one region is selected from the plurality of divided regions. Then, analysis processing of the image data is performed using the histogram distribution included in the selected area as the data to be analyzed. Therefore, by selecting an effective image area to be reproduced from the image data as the one area, the analyzed data can be limited to a histogram distribution representing the effective image area. Therefore, even when the exposure state of the image is in an inappropriate state such as “backlight” or “under exposure”, it is possible to perform proper image analysis using the histogram distribution of the effective image area portion as the analyzed data. it can.

In the above-mentioned image analysis method, the one area selected in the step (c) may be configured to be an area at least representing a subject which is a subject in the image.

According to this configuration, it is possible to properly analyze a subject as a subject in an image.

In the above-structured image analysis method, the step (c) includes: (c1) displaying the image data on a display screen; and (c2) receiving designation of a position on the screen by an operator. Selecting arbitrary pixel data from the image data displayed on the screen; (c3)
Selecting the one area including the selected pixel data.

According to this configuration, the operator can designate a valid area portion in the image. Therefore, it is possible to appropriately analyze the image target intended by the operator.

In the above-structured image analysis method, (e)
(F) determining the exposure state of the image; and (f) when the determined exposure state is backlight,
Are sequentially performed.

According to this configuration, when the exposure state of the image is backlight, appropriate image analysis is possible.

An image analyzing apparatus according to the present invention is an image analyzing apparatus for performing analysis processing of image data using a histogram of image data representing an image as analyzed data, wherein a histogram distribution has at least two peaks. Histogram storing means for storing a histogram stored in the histogram storing means, a histogram dividing means for dividing the histogram stored in the histogram storing means into a plurality of areas for each of the peaks, and one of the plurality of divided areas. The gist of the invention is to provide an area selecting means for selecting an area and a histogram distribution extracting means for extracting a histogram distribution included in the selected area as the analyzed data.

The image analysis apparatus having this configuration has the same operation and effect as the above-described image analysis method, and can perform appropriate image analysis regardless of the exposure state of an image.

In this image analyzing apparatus, the one area selected by the area selecting means may be an area representing at least a subject which is a subject in the image.

According to this configuration, similarly to the method of the present invention (the image analysis method according to the second aspect), it is possible to appropriately analyze a subject which is a subject in an image.

[0019]

Other Embodiments of the Invention The present invention includes the following other embodiments. First, a first aspect is a portable storage medium that stores a computer program which is executed by a microprocessor of a computer to realize each step or each means of the above invention. Also, the second
Is an aspect as a program supply device that supplies the computer program via a communication path.
In the second aspect, a method for extracting a reference density gradation value of an image by placing a program on a server or the like on a network, downloading a necessary program to a computer via a communication path, and executing the program. And a device can be realized.

[0020]

Next, embodiments of the present invention will be described based on examples.

FIG. 1 is a block diagram showing an image data creating apparatus to which one embodiment of the present invention is applied. This image data creating apparatus is an apparatus for obtaining image data for creating a printing plate from an original such as a photographic original, and an image input apparatus 10 for reading an original G and obtaining multi-gradation image data.
And an image processing device 20 that analyzes the multi-tone image data obtained by the image input device 10 and performs various data processing on the multi-tone image data.

An image input device 10 optically reads an original G and outputs a CCD element 1 which outputs analog image signals representing three primary color components of R (red), G (green) and B (blue).
1, a magnification changing circuit 12 having a function of converting an output signal of the CCD element 11 into a digital signal and a function of changing magnification,
A gamma correction circuit 13 for gamma correction is provided.

The magnification changing circuit 12 can set a plurality of types of sampling periods when converting an analog image signal into digital image data, and reads an image depending on which sampling period is applied. The magnification is determined. For example, if the sampling period is determined so that the output signal of one read pixel of the CCD element 11 is sampled twice, the original image can be enlarged twice.

The gamma correction circuit 13 performs data correction for compensating for gamma characteristics caused by the characteristics of the CCD element 11 and creates multi-tone image data that faithfully reproduces the original G. The RGB three-primary-color multi-tone image data (hereinafter, simply referred to as “image data”) output from the image input device 10 is taken into the image processing device 20 as an image file F.

The image processing device 20 is composed of, for example, a personal computer, and includes an external storage device 40, an input device 50, and a display device 60 as peripheral devices around the main control unit 30. The external storage device 40 is configured by a hard disk unit that temporarily stores various data, and stores an image file F from the image input device 10.
And a reference database DB described later. The input device 50 includes a mouse 52 and a keyboard (not shown). The display device 60 is configured by a CRT display.

The main control unit 30 executes a software program (computer program) stored in an internal storage device (not shown) by a CPU (not shown) to thereby execute a density gradation conversion unit 31, a color conversion unit 32, and a histogram creation unit 33. , Exposure state determination unit 34, subject target specification unit 3
5. The functions as the histogram division unit 36, the histogram distribution extraction unit 37, and the LUT creation unit 38 are realized.
Note that the histogram creation unit 33 stores the histogram division unit 3 in the histogram storage unit of the image analysis device of the present invention.
Reference numeral 6 corresponds to a histogram distribution unit, and the histogram distribution extraction unit 37 corresponds to a histogram distribution extraction unit. Further, the LUT creation unit 38 corresponds to the function of executing an analysis process based on the analyzed data in the present invention.

The software programs for realizing the above functions are provided in a form recorded on a computer-readable recording medium such as a floppy disk, CD-ROM, magneto-optical disk, IC card, or the like. The computer reads the software program from the recording medium and transfers it to the internal storage device or the external storage device 40.
Alternatively, a software program may be supplied to a computer via a communication path. Specifically, the computer is configured to be connected to a modem, the modem is connected to a communication line, and the communication line is connected to a network including a server. Accordingly, the server has a function as a program supply device that supplies the software program to the image processing device 20 via the communication line.

When realizing the functions of the software program, the software program stored in the internal storage device is executed by the CPU. Further, the computer may read the software program recorded on the recording medium and directly execute the program by the CPU.

Each unit 31 to 31 realized by the main control unit 30
38 will be described in detail below. Here, the flow of the entire processing by each of the units 31 to 38 will be described first.

The image data d1 of the image file F output from the image input device 10 is received by the density gradation conversion section 31, and the density gradation conversion processing for distributing the density distribution biased with respect to the image data d1 is performed. Will be applied. This density gradation conversion process is performed according to a look-up table (hereinafter, referred to as an LUT) created by the LUT creating unit 38. The image data after the density gradation conversion is color-converted into C (cyan), M (magenta), Y (yellow) and K (black) image data by the color conversion unit 32, and is externally provided as plate-making image data d2. Is output. The plate making image data d2 is used in an image recording apparatus or the like that creates a printing plate by exposing a photosensitive material film.

The image data d1 of the image file F is also sent to the histogram creating section 33, which creates a histogram for the density gradation value of the image data d1. The created histogram is sent to the exposure state determination unit 34. The exposure state determination unit 34 determines the exposure state of the image data d1 from the histogram of the image data d1 while referring to the reference database DB.

A signal indicating the determined exposure state is sent to the histogram division unit 36. Histogram division unit 3
In step 6, the process of dividing the histogram created by the histogram creating unit 33 into a plurality of regions is performed. Whether or not this process is necessary is determined according to the exposure state. The subject target specifying unit 35 fetches the image data d1 of the image file F and performs a process of specifying a subject as a subject in the image data d1. Here, the subject as a subject refers to a subject such as a person who wants to reproduce the image data d1 in an appropriate exposure state.

The signal indicating the subject obtained by the subject specifying section 35 and the signal indicating a plurality of divided areas obtained by the histogram dividing section 36 are sent to a histogram distribution extracting section 37. The histogram distribution extracting unit 37 performs a process of selecting one divided region including the subject from the plurality of divided regions and extracting a histogram distribution of the selected divided region. This histogram distribution is sent to the LUT creating unit 38 as analyzed data, and the LUT creating unit 38 creates a lookup table (LUT) for density gradation conversion based on the analyzed data. This LUT is sent to the density gradation conversion unit 31, and is used for density gradation conversion processing.

Next, the units 31 to 38 realized by the main control unit 30 will be described in detail in order.

The histogram creating unit 33 creates a histogram of the image data d1 of the image file F, as described above. As shown in FIGS. 2 and 3, the histogram created here is data representing a stepwise graph in which the horizontal axis represents the density gradation value of the image and the vertical axis represents the number of appearances. Note that the density gradation value on the horizontal axis increases as it approaches the origin side.

FIGS. 2 and 3 are diagrams of histograms for various images. In FIG. 2, (a) to (d)
FIG. 3 shows some examples of histograms for an image determined to be “appropriate exposure”.
(D) show some examples of histograms for images determined to be "underexposed".

As shown in FIGS. 2A to 2D,
Even in the same “appropriate exposure” state, various histogram distributions are obtained depending on the photographed situation and the type of the subject, and as shown in FIGS. Even in the state, various histogram distributions are taken depending on the degree of light quantity shortage, the shooting situation, and the like.

As described above, there are various patterns in the shape of the histogram distribution depending on the target image data. The exposure state determination unit 34 uses a neural network in which a characteristic histogram shape is learned in advance as an input pattern, thereby expecting a reasonable output even for a pattern that does not completely match the pattern used for learning. can do. In particular, a neural network has a parallel processing function of a human perception system, and is therefore suitable for non-quantitative recognition of a histogram-shaped pattern to be recognized. Therefore, a neural network is applied to the exposure state determination unit 34 of this embodiment.

FIG. 4 is a block diagram for explaining the contents of processing in the exposure state determination section 34. As shown in the figure, the exposure state determination unit 34 includes a normalization processing unit 34a for normalizing the histogram and basic information for determining the exposure state of the target image by performing pattern recognition of the normalized histogram. Output neural network 3
4b.

The normalization processing section 34a eliminates the influence of the total number of pixels of the target image. In other words, the greater the total number of pixels of the target image, the higher the frequency in the histogram. Accordingly, for shape analysis, it is preferable to eliminate the influence of the total number of pixels. Therefore, the normalization processing unit 3
4a calculates an average value of frequency values corresponding to response values within a certain range near the response value having the maximum frequency in the histogram created by the histogram creating unit 33, and sets the average value to a predetermined value. Change the frequency value corresponding to each response value. Thereby, normalization of the histogram is achieved. The reason for taking the average value is that if the histogram is comb-shaped, normalization cannot be performed properly even if the frequency value of each response value is changed so that the maximum frequency value matches the predetermined value. is there.

When determining the response value that takes the maximum frequency,
It is preferable that response values near “0” be excluded from candidates. This is because the frequency near the response value “0” is expected to be extremely large. Therefore, even if the normalization processing is performed based on the maximum frequency value corresponding to the response value near the response value “0”, This is because proper normalization may not be performed.

FIG. 5 is a diagram conceptually showing the structure of the neural network 34b. This neural network is a feed-forward type nonlinear element network called a multilayer perceptron. This neural network has three layers: an input layer, a hidden layer, and an output layer.
Consists of layers. Of course, those having a larger number of layers are also applicable.

"O" in FIG. 5 represents a non-linear element. For example, assuming that the input / output condition of the i-element in the figure is X for the input vector from the previous layer and Y for the output vector of the element, It can be defined in 1).

[0044]

(Equation 1)

The function f is a monotonically increasing function for giving a non-linear characteristic. Generally, a sigmoid function whose output monotonically increases within the range of (0, 1) is used. The sigmoid function is generally defined by the following equation (2).

[0046]

(Equation 2)

FIG. 6 is a diagram for explaining an input pattern to the nonlinear element of the input layer, and shows a normalized histogram. Each nonlinear element in the input layer of the neural network 34b has sampling points S1, S2,
.., The frequency values (normalized values) in S10 are input.

In FIG. 6, the sampling point S
1, S2, ..., S10 were determined at equal intervals,
The intervals need not always be equal. In determining the sampling point, it is possible to adopt a configuration in which an optimum response value is selected to detect a change in the shape of the histogram distribution depending on the magnitude of the density value. More specifically, a sampling point is set at a high density in the vicinity of a small response value or in the vicinity of a sudden change in the rate of change in frequency (slope of the histogram) with respect to the response value. On the other hand, when the response value is near a relatively large value, it is considered that the change in the frequency due to the magnitude of the noise amount is small, and thus the sampling points are set at wide intervals.

The number of samplings is arbitrary, but if the number is too large, the number of nonlinear elements must be increased accordingly, which leads to an increase in calculation time. Therefore, it is preferable to keep the number as small as possible. Further, in order to reduce the influence of the local fluctuation of the distribution, the histogram is subjected to a smoothing process in the vicinity of the sampling point, and the smoothed value is used as sampling data in the neural network 34.
What is necessary is just to input to the nonlinear element of the input layer of b.

For example, when the number of samplings is five,
When an output pattern is classified into four categories of “appropriate exposure”, “under exposure”, “over exposure”, and “backlight” with respect to an input pattern, a perceptron model is as shown in FIG.

That is, the input layer has five nonlinear elements corresponding to the number of samplings, and sample data D1, D2,..., D5 are input to these five nonlinear elements. The input layer is further provided with a threshold bias element.

The intermediate layer includes three non-linear elements and one threshold bias element. The outputs of the non-linear elements of the input layer and the outputs of the bias elements are given in parallel to each of the three non-linear elements. The outputs of the three non-linear elements in the intermediate layer and one threshold bias element are input in parallel to each of the four non-linear elements in the output layer. The four non-linear elements of the output layer output signals indicating the degree of belonging to each of the categories of “appropriate exposure”, “under exposure”, “over exposure”, and “backlight”. This allows
It can be said that the neural network 34b constitutes an exposure state determination unit.

In order for the neural network 34b to perform appropriate pattern recognition of the exposure state, first, learning must be performed on the defined perceptron. The learning method is, for example, to prepare a set of images previously classified into four categories (“appropriate exposure”, “underexposure”, “overexposure”, and “backlight”) and use them as a teacher to perform a reverse transfer algorithm. It should be done by. The procedure is described below.

A normalized histogram is created from the images of each category. The normalized histogram is sampled to obtain a perceptron input pattern. A default (default) combination weighting coefficient is given to the perceptron, and the pattern obtained by is input and an output value is obtained.

With the output pattern of the category of the input image as the output target, the Euclidean distance from the output pattern of the category is calculated. In order to fill the Euclidean distance, a new joint weighting coefficient is calculated by the inverse transfer algorithm. Until the Euclidean distance between the output target and the output pattern of the input image converges to a constant value, a new combination weighting coefficient is determined to obtain an output pattern, and further the Euclidean distance between this output pattern and the output target is calculated. Repeat steps.

The perceptron learned in this manner is expected to output a valid classification result even for an input pattern corresponding to an unspecified image other than the image used as the teacher. Also, even if the classification cannot be clearly determined by the interpolation function of the neural network itself,
An intermediate output pattern can be expected. Thus, the neural network 34b outputs information indicating the exposure state of the target image.

The reference addition database DB referred to by the exposure state determination section 34 stores the combined addition coefficient determined by the learning process. The exposure state determination result is fed back to the reference database DB as needed. That is, when the density gradation conversion processing by the density gradation conversion unit 31 is not properly performed, learning is performed again with the category of the image at that time as an output target, so that the combination weighting coefficient is updated. Thus, as the number of times of image processing increases, the accuracy of the determination of the exposure state increases.

Next, a description will be given of the subject target specifying unit 35, the histogram dividing unit 36, and the histogram distribution extracting unit 37. According to these three parts 35,36,37,
As described above, the subject which is the subject in the image data d1 is specified, the histogram created by the histogram creating unit 33 is divided into a plurality of divided areas, and one of the divided areas includes the subject. A divided region is selected, and a histogram distribution for the selected divided region is extracted. The three units 35, 36, and 37 are realized by the CPU executing a software program that defines a series of these processes (hereinafter, referred to as an analyzed data generation process).

FIG. 8 is a flowchart showing the analyzed data generation processing routine executed by the CPU. As shown, when the process is started, the CPU first performs a process of inputting a histogram created by the histogram creating unit 33 and an output pattern of the neural network by the exposure state determining unit 34 (step S100,
S110). Next, it is determined whether or not the input output pattern indicates a “backlit” exposure state (step S120). Here, if it is determined that it is not “backlit”, the histogram input in step S100 is stored as analyzed data (step S130), and then the process exits to “return” and ends the process.

On the other hand, if it is determined in step S120 that the subject is in the "backlit" exposure state, the subject target designation processing (step S200) and the histogram division processing (step S30)
0), histogram distribution extraction processing (step S400)
In order. Thereafter, the process exits to "return" and ends the process.

FIG. 9 is a flowchart showing the subject object designation processing in step S200. This subject object designation process is a subroutine call from step S200, and is started by the CPU.

As shown, when the process is started, C
The PU first converts the image data d1 of the image file F into a CR.
It is displayed on the T display 60 (step S210).
Next, a process of designating a subject as a subject from the displayed image on the CRT display 60 is performed (step S2).
20). More specifically, in response to an operation of the mouse 52 by an operator, the position on the image pointed by the mouse 52 is determined as the subject. As shown in FIG. 10, an image of the image data d1 is displayed on the CRT display 60, and when the mouse 52 is operated, a subject in the image, for example, a pointer PT is set to a person PP, so that the subject is displayed. The target is determined.

When the subject is specified in step S220, the CPU then performs a process of extracting the density gradation value of the unit pixel on the image data d1 corresponding to the specified position (step S230). Thereafter, the process returns to "return" and the subject object designation processing is temporarily ended.

FIG. 11 is a flowchart showing the histogram division processing in step S300. This histogram division process is a subroutine call from step S300 and is started by the CPU.

As shown, when the process is started, C
The PU first analyzes the histogram input in step S100, and performs a process of searching for vertices of two peaks in the histogram distribution (step S310).
The histogram used for this processing is obtained in step S120.
Is limited to images in the "backlit" exposure state. The histogram of the image in the “backlit” exposure state has a bimodal shape having two peaks M1 and M2, for example, as shown in FIG. Here, the peak M2 on the side where the density gradation value is low (the brightness is high) is an area that is illuminated more than necessary in the backlight exposure state and does not need to be reproduced in an appropriate exposure state. In step S310, a process of searching for the vertices P1 and P2 of the peaks M1 and M2 from such a histogram is performed.

Next, a process of dividing the area between the vertices P1 and P2 into a plurality of areas having a predetermined interval is performed in the horizontal axis direction of the histogram (step S320). FIG.
Is a diagram for explaining the state of the division and the like. FIG.
(A), a distance d between the vertices P1 and P2 is set to an appropriate size so as not to be affected by the local saddle portion.
, En (n is a positive number) having a plurality of regions E1, E2,.

Subsequently, each of the divided areas E1, E2,.
, The average value of the number of appearances is calculated (step S330). The graph obtained by plotting the average values thus obtained is shown in FIG. Subsequently, a divided area having the minimum average value is selected from among the divided areas E1, E2,..., En (step S340).

When the divided area having the minimum average value is obtained,
Next, the centers of the minimum divided areas are defined as peaks P1 and P2.
It is determined as a threshold value Th for division between (Step S35)
0). Here, as shown in FIG. 13B, when there are a plurality of minimum divided areas, the center of the section (minimum section) extending over the plurality of divided areas is determined as the threshold Th. Thereafter, a process of dividing the histogram used for this process into two regions (first region and second region) by the threshold Th is performed (step S360). That is, as shown in FIG. 14, the histogram is divided into the first area H1 and the second area H2 by the threshold Th.

Thereafter, the process returns to the "return" and the histogram division processing is temporarily terminated.

FIG. 15 is a flowchart showing the histogram distribution extracting process in step S400. This histogram distribution extraction processing is a subroutine called from step S400 and is started by the CPU.

As shown, when the process is started, C
First, the PU performs a process of selecting, from the first region H1 and the second region H2 obtained by the histogram division process, a side including the density gradation value extracted by the subject target designation process (step S410). Subsequently, a process of extracting a histogram distribution included in the selected area (H1 or H2) is performed (step S420). For example, as shown in FIG. 16, assuming that the density gradation value extracted by the subject target designation processing is the part indicated by the arrow X in the figure
The first area H1 on the side including the density gradation value is selected in step S410, and in step S420,
The histogram distribution included in the selected first region H1 is extracted.

The histogram distribution extracted in step S420 is stored as analyzed data (step S43).
0) Then, the process returns to “return” and the histogram distribution extraction processing is temporarily terminated.

Next, the LUT creation section 38 will be described.
The LUT creation unit 38 creates a look-up table (LUT) for density gradation conversion based on the analyzed data obtained by the analyzed data generation processing routine.
The software program that defines the LUT creation process is CP
This is realized by U executing.

FIG. 17 is a flowchart showing the LUT creation processing executed by the CPU. As shown in the figure, when the process is started, the CPU first analyzes the analyzed data obtained by the analyzed data generation process routine (that is, when the exposure state is backlight, the histogram extracted by the histogram distribution extraction process). If the exposure state is other than backlight, the process is to create a cumulative histogram from the histogram (the histogram created by the histogram creating unit) (step S500). As shown in FIG. 18, the cumulative histogram is a graph in which the frequencies of appearance with respect to the density gradation values shown on the horizontal axis are sequentially accumulated, and the cumulative histogram is shown on the vertical axis.

Then, as seen from the distribution of the cumulative histogram, as shown in FIG.
[%] Of the first density gradation value D1 and 50
A second density gradation value D2 corresponding to [%] is obtained (step S510).

The first density gradation value D1 obtained here
Is a density tone at a predetermined position near the tone value at the boundary where the cumulative frequency distribution is maximum, and is a so-called density tone value at a highlight point. In this embodiment, the value of the first density gradation value D1 is set at a point where the cumulative value of the appearance frequency becomes 99.5 [%], but in the vicinity of the point where the cumulative value becomes 100 [%]. May be a point of 99.0 [%] or a point of 98.0 [%].

Subsequently, the first and second density gradation values D1 and D2 obtained in step S510 and 99.5 [%],
Based on the cumulative appearance frequency of 50%, a process of creating an LUT for density gradation conversion is performed (step S2).
50). Specifically, as shown in FIG. 19, a two-dimensional coordinate axis is prepared in which the horizontal axis is set to the input density and the vertical axis is set to the output density, and the first and second density gradation values D1, D1 are set on the horizontal axis. D
2 is a value (25) corresponding to the above 99.5 [%] on the vertical axis.
95.5 [%] of 6 gradations, such as "255")
And a value corresponding to 50 [%] (25 represented by 8 bits)
A value such as “128” which is 50 [%] of six gradations) is taken. Then, the first and second density gradation values D on the horizontal axis
A curve having input / output characteristics corresponding to values 255, 128 on the vertical axis corresponding to 1, D2 is drawn on the coordinate axis. Thus,
Create an LUT.

Thereafter, the CPU advances the processing to "return", and ends the processing routine once.

According to the LUT creation processing having such a configuration,
An LUT that can realize density gradation conversion for distributing a biased distribution of density gradation values throughout is created.

As described in detail above, according to the image processing apparatus 20 of this embodiment, the exposure state of the image of the original G is determined, and if the determined exposure state is "backlit", the image is From the histogram, a divided region including a peak portion representing a subject to be a subject in the image is extracted, and a histogram distribution in the extracted divided region is defined as analyzed data.

According to the conventional technique, in an image whose exposure state is "backlit", an invalid density gradation area (for example, an area of a peak M2 shown in FIG. 12) caused by the exposure state is included in the analyzed data. Therefore, the highlight point is determined within the invalid density gradation range. On the other hand, according to this embodiment, since only the histogram distribution included in the divided region of the peak portion representing the target object in the image is the analyzed data, the peak portion of the subject object is divided. Highlight points can be detected in the area. Therefore, even in the “backlit” exposure state, it is possible to obtain an appropriate density gradation value at a highlight point, and to perform appropriate image analysis. As a result, it is possible to perform appropriate density gradation conversion processing.

Further, in this embodiment, since the operator can select the divided area including the object to be the subject in the image by operating the mouse 52, the image object intended by the operator can be appropriately selected. This also has an effect that the analysis can be performed in a short time.

In this embodiment, as described above,
The divided area including the target object in the image is
2, the operator can select the high density side divided area H1 automatically without the operator's instruction. This is because, when the exposure state of the image is “backlit”, since the subject of the image is generally included in the peak on the high density side, the divided region H1 on the high density side is included in the subject of the image. For example, it is possible to adopt a configuration in which the high-density side divided area is automatically selected without an instruction from the operator.

Further, in this embodiment, the case where the exposure state of the image is "backlight" has been described. Instead, the judgment in step S120 of the analyzed data generation processing routine is added to "backlight" and "underlight". In the case of "exposure", the histogram distribution may be extracted. This is because, similarly to the image in the “backlight” exposure state, the histogram of the “underexposed” exposure state shows a bimodal shape, and the high-density side divided region includes the subject of the image. This is because it is common.

According to this configuration, it is possible to obtain an appropriate density gradation value of a highlight point in the case of the "under exposure" exposure state in addition to the "backlight" exposure state. An analysis can be performed.

In the above embodiment, a bimodal histogram having two peaks is prepared as an image histogram.
Although the peak side divided region including the region representing the subject which is the subject of the image is selected, a histogram having a distribution having three or more peaks may be prepared instead. . Also in this configuration, a configuration is adopted in which, when an operator designates a region representing a target which is a subject of an image, a divided region including a peak including the region is selected.

With this configuration, it is possible to perform appropriate image analysis on images in various exposure states.

In this embodiment, image analysis is performed such that a histogram in an appropriate range is used as analyzed data and an LUT for returning a density gradation is created using the analyzed data. The analysis does not need to be limited to the process of creating the LUT for returning the density gradation. For example, a configuration in which the processing parameter for executing the image enhancement process for enhancing the density gradation value is obtained using the data to be analyzed. It may be.

Further, the following modifications may be made.
Instead of the method of calculating the threshold value Th for division in the histogram division unit 36 described in the above embodiment, a configuration in which the threshold value Th is calculated by the following method may be adopted.

That is, the method executes the following steps in order. Sampling is appropriately performed between the vertices P1 and P2. In that section, a cubic equation is approximated using the least squares method. The obtained cubic expression is differentiated, and the x coordinate giving the extreme value is determined as the threshold Th.

FIG. 20 shows a method based on the above least square method.
FIG. 9 is an explanatory diagram in a case where the sampling is performed using all sampling points. As shown, since the cubic expression U can be approximated with high accuracy by using all the sampling points, the threshold value Th with high accuracy can be obtained.

FIG. 21 shows a method using the above least squares method.
FIG. 9 is an explanatory diagram in a case where the sampling is performed using limited sampling points. As shown in the figure, even if the limited sampling points are used, the cubic equation U can be approximated with a certain degree of accuracy, so that a small number of sampling points are sufficient and high-speed processing can be performed.

The embodiments of the present invention have been described above.
The present invention is not limited to such embodiments at all, and it goes without saying that the present invention can be implemented in various modes without departing from the gist of the present invention.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an image data creating apparatus to which an embodiment of the present invention is applied.

FIG. 2 is an explanatory diagram illustrating an example of a histogram of density gradation for an image of “appropriate exposure”.

FIG. 3 is an explanatory diagram illustrating an example of a histogram of density gradation for an “under-exposed” image.

FIG. 4 is a block diagram for explaining processing contents in an exposure state determination unit 34;

FIG. 5 is an explanatory diagram conceptually showing the structure of a neural network 34b.

FIG. 6 is a graph showing a normalized histogram for explaining an input pattern to a nonlinear element of an input layer.

FIG. 7 is an explanatory diagram showing an example of a neural network.

FIG. 8 is a flowchart illustrating an analyzed data generation processing routine executed by a CPU.

FIG. 9 is a flowchart showing a subject object designation process in step S200.

10 shows a CRT display 60 using a mouse 52. FIG.
FIG. 9 is an explanatory diagram showing a state in which a subject of a subject is designated from among the images.

FIG. 11 is a flowchart illustrating a histogram division process in step S300.

FIG. 12 is a graph showing an example of a histogram of an image in a “backlit” exposure state.

FIG. 13 is an explanatory diagram for explaining a method of dividing a region between the vertex P1 and the vertex P2 into a plurality of regions and dividing the region between the vertex P1 and the vertex P2 into two in the histogram.

FIG. 14 is an explanatory diagram showing a first area H1 and a second area H2 divided by the above method.

FIG. 15 is a flowchart showing a histogram distribution extraction process in step S400.

FIG. 16 is an explanatory diagram illustrating a method of selecting one region H1.

FIG. 17 is a flowchart illustrating an LUT creation process executed by the CPU.

FIG. 18 is a graph showing a cumulative histogram.

FIG. 19 shows an L for density gradation conversion created in this embodiment.
It is a graph which shows UT.

FIG. 20 is an explanatory diagram for describing a method of obtaining a threshold using the least squares method after using all sampling points.

FIG. 21 is an explanatory diagram for explaining a method of obtaining a threshold using a least squares method after using limited sampling points.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 ... Image input device 11 ... CCD element 12 ... Magnification change circuit 13 ... Gamma correction circuit 20 ... Image processing device 30 ... Main control part 31 ... Density gradation conversion part 32 ... Color conversion part 33 ... Histogram creation part 34 ... Exposure state Discriminating unit 34a Normalization processing unit 34b Neural network 35 Subject designation unit 36 Histogram division unit 37 Histogram distribution extraction unit 38 LUT creation unit 40 External storage device 50 Input device 52 Mouse 60 CRT display H1: first area H2: second area Th: threshold

Claims (6)

[Claims] 1. An image analysis method for performing analysis processing of image data using a histogram of image data representing an image as analyzed data, wherein (a) a histogram distribution has at least two peaks. Preparing a histogram, and (b) the step (a)
Dividing the histogram prepared in the above into a plurality of regions for each peak; (c) selecting one region from the plurality of divided regions; and (d) selecting one of the divided regions. Extracting the included histogram distribution as the analyzed data. 2. The image analysis method according to claim 1, wherein the one area selected in the step (c) is an area representing at least a subject which is a subject in the image. 3. The image analysis method according to claim 1, wherein the step (c) comprises: (c1) displaying the image data on a screen of a display; and (c2) displaying the image data on the screen by an operator. Receiving the designation of the position of, select any pixel data of the image data displayed on the screen,
(C3) selecting the one area including the selected pixel data. 4. The image analysis method according to claim 1, wherein (e) a step of determining an exposure state of the image, and (f) when the determined exposure state is a backlight. And (c) sequentially executing steps (b) to (d). 5. An image analysis apparatus for performing analysis processing of image data, using a histogram of image data representing an image as analyzed data, storing a histogram whose histogram distribution has at least two peaks. A histogram storage means for performing the processing, and a histogram stored in the histogram storage means.
A histogram dividing unit that divides each peak into a plurality of regions; a region selecting unit that selects one region from the plurality of divided regions; and a histogram distribution included in the selected region, An image analysis apparatus comprising: a histogram distribution extraction unit that extracts the data as analysis data. 6. The image analysis apparatus according to claim 5, wherein the one area selected by the area selection means is an area that at least represents a subject as a subject in the image.