JP2008027058A - Image detection apparatus and image detection method - Google Patents

Abstract

To improve detection accuracy when a face image detection device detects a face image.
A face image detection area classification unit 133 determines a reference area 200A and a determination area 200B having relative indices classified as an area 13A as face image detection areas of the same person, and is classified as an area 13B. The reference area 200A and the identified area 200B having the relative indices are determined as the face image detection areas of different persons, and the reference area 200A and the identified area 200B having the relative indices classified into the areas 13C and 13D are identified. The discriminated area 200B is discriminated as an erroneous detection area.
[Selection] Figure 12

Description

  The present invention relates to an image detection apparatus and method for detecting a predetermined image from an image.

  A technique for detecting a predetermined image such as a face image from an input image is used as, for example, pre-processing for determining the presence or absence of a face, identifying a face, and detecting a face part.

  Some face image detection devices that detect a face image from an input image, for example, detect a plurality of areas around the original face image area. In general, such a face image detection device uses a method of selecting a detection result having a maximum predetermined score value from among a plurality of detection results. For example, a search width when searching for a face image, The selected detection result may be a discrete value due to the fact that the reduction ratio of the image is a discrete value.

  Therefore, for example, as shown in Patent Document 1 below, there is a method of stabilizing a face image area detected using an average value of each scale in a plurality of images having different resolutions.

JP 2006-48328 A

  However, in the technique described in Patent Document 1, the detection result is obtained very close to the original face image area when the face areas of two or more persons are very close to each other or due to erroneous detection. May end up. In such a technique, unless the face detection area used for the calculation for stabilizing the face image area is properly selected from the plurality of face detection areas, the detection accuracy is greatly reduced.

  The present invention has been proposed in view of such a conventional situation, and an image detection apparatus and method for appropriately selecting a face detection area used for calculation for stabilizing a detected face image area The purpose is to provide.

  In order to achieve the above-described object, the present invention provides an image detection apparatus that detects a predetermined image from an input image, from the input image every time the relative size between the detection target area of the predetermined size and the input image is changed. Image discriminating means that scans and cuts out the detection target area so as to move by predetermined pixels, determines whether or not the image in the cut out detection target area is the predetermined image, and outputs a discrimination result And when the image discriminating means discriminates that the image in the detection target area is the predetermined image, the relative relationship between the two detection areas including the reference area and the discrimination area detected for the predetermined image is determined. Index value calculation means for calculating a predetermined index value based on the position and relative size, and classification based on the predetermined index value calculated by the index value calculation means Serial to be discrimination region, characterized in that it comprises a detection area determining means for determining whether or not to adopt the calculation process for generating a detection result of the predetermined image.

  In order to achieve the above-described object, the present invention provides an image detection method for detecting a predetermined image from an input image every time the relative size between a detection target region of a predetermined size and the input image is changed. An image that scans and cuts out the detection target area from the image so as to move by predetermined pixels, determines whether or not the image in the cut out detection target area is the predetermined image, and outputs a determination result If the image in the detection target region is determined to be the predetermined image in the determination step and the image determination step, two detections including a reference region and a determination target region detected for the predetermined image An index value calculation step for calculating a predetermined index value based on the relative position and relative size of the region, and a classification based on the predetermined index value calculated in the index value calculation step To be discrimination area, and having a detection region determination step of determining whether to employ the calculation process for generating a detection result of the predetermined image.

  The present invention relates to an image detection apparatus and method for detecting a predetermined image from an input image, and the detection target region is determined from the input image every time the relative size between the detection target region of a predetermined size and the input image is changed. The image is scanned and moved so as to move each pixel, and it is determined whether or not the image in the detected detection area is the predetermined image, and the determination result is output. The image in the detection target area is When it is determined that the image is the predetermined image, a predetermined index value is calculated based on the relative position and the relative size of the two detection regions including the reference region and the determination target region detected for the predetermined image, It is discriminated whether or not the discriminated area classified based on the calculated predetermined index value is adopted in a calculation process for generating a detection result of the predetermined image. And it makes it possible to further stabilize the detection area of a predetermined image as compared with the conventional image sensing apparatus and method.

  Hereinafter, specific embodiments to which the present invention is applied will be described in detail with reference to the drawings.

  FIG. 1 is a diagram showing a face image detection apparatus 1 according to an embodiment to which the present invention is applied. The face image detection apparatus 1 detects a face image from an image based on input image data, and outputs a detection result of a face image detection area or the like.

  The face image detection device 1 is configured as a personal computer having the internal configuration shown in FIG.

  As shown in FIG. 2, the face image detection device 1 receives image data from an operation unit 11 to which user operation information is input via, for example, a keyboard and a mouse, and from a digital camera, a television device, a storage, and the like. The image input unit 12, the face image detection unit 13 for detecting a face image from the image based on the input image data, the recording unit 14 for recording the image data, and performing predetermined image processing on the display A face image detection device 1 including a buffer memory, a RAM (Random Access Memory), a ROM (Read Only Memory) in which a face image detection program is recorded, a CPU (Central Processing Unit), and the like. The control unit 16 for controlling the overall signal processing is connected to the internal bus 17.

  As shown in FIG. 3, the face image detection unit 13 includes a face image determination unit 131, a determination index value calculation unit 132, a face image detection region classification unit 133, and a face image region averaging unit 134. Is done.

  When image data is input via the image input unit 12 based on the control operation of the control unit 16, the face image determination unit 131 converts a face image in an 80 × 100 pixel image 100 </ b> A based on the input image data. A rectangular area 101 of 20 × 20 pixels for discrimination is moved rightward or downward by two pixels from the upper left end of the image 100A shown in FIG. 4A, and the right side of the image 100A shown in FIG. Let reach the bottom. The face image determination unit 131 scans and cuts out an image of the rectangular area 101 every time the rectangular area 101 is moved by two pixels, and determines whether or not a predetermined image existing in the rectangular area 101 is a face image. To do.

  That is, the face image discriminating unit 131 is set in advance as an image existing in the rectangular area 101 by a method such as support vector machine (SVM), boosting such as Adaboost, or local feature amount. By comparing the size and the degree of correlation with the face image, it is determined whether or not the predetermined image existing in the rectangular area 101 is a face image.

Then, the face image determination unit 131 calculates a score value S _i indicating the degree of correlation with a preset face image. If the score value S _i is equal to or greater than a certain threshold value, the face image determination unit 131 stores the score value S _i in the rectangular area 101. It is determined that the existing image is a face image.

The face image discriminating unit 131 performs scanning and clipping processing while moving the rectangular region 101 from the upper left end of the image 100A shown in FIG. 4A to the lower right end of the image 100A shown in FIG. As shown in FIG. 4B, when it is determined that the predetermined image existing in the rectangular area 101 is a face image, the xy coordinate values x _i , y _i at the upper left end of the rectangular area 101, the face image detection area The width w _i and the face image detection area height h _i are calculated, and the xy coordinate values x _i and y _i , the face image detection area width w _i , and the face image detection area height h _i are obtained as a discrimination result. The score value S _i is supplied to the subsequent discrimination index value calculation unit 132.

Note that the score value S _i, may also be higher score values S _i is close to the shape of the face of the face image detected common person shape is preset is calculated, the calculation of the other It may be calculated by a method.

  On the other hand, when the face image determining unit 131 determines that the predetermined image existing in the rectangular area 101 is not a face image, the face image determining unit 131 moves the rectangular area 101 by two pixels without performing the above-described calculation process. Face image discrimination processing is performed.

  The face image discriminating unit 131 moves the rectangular area 101 to the lower right end of the image 100A in the image 100A, and then reduces the image 100A at a predetermined magnification as shown in FIG. 100B.) In the image 100B, the rectangular region 101 is moved each time the 20 × 20 pixel rectangular region 101 is moved rightward or downward by two pixels from the upper left end of the image 100B, as in the case of the image 100A. Is cut out to determine whether or not the predetermined image existing in the rectangular area 101 is a face image.

As shown in FIG. 4E, the face image discriminating unit 131 moves the rectangular area 101 to the lower right end on the image 100B, and then the predetermined image existing in the rectangular area 101 is faced at any position. When it is determined that the image is not an image, as shown in FIG. 4F, the image 100B is reduced by a predetermined magnification (this image is referred to as an image 100C), and the processing in the images 100A and 100B is performed on the image 100C. In the same manner as above, each time the rectangular area 101 of 20 × 20 pixels is moved rightward or downward by two pixels from the upper left end to the lower right end on the image 100C, the rectangular area 101 is scanned and cut out. It is determined by calculating the score value S _i whether or not the existing image is a face image.

The face image discriminating unit 131 performs scanning and clipping processing while moving the rectangular area 101 from the upper left end of the image 100C shown in FIG. 4F to the lower right end of the image 100C shown in FIG. 4 (G), when it is determined that the predetermined image existing in the rectangular area 101 is a face image, the xy coordinate values x _i , y _i at the upper left end of the rectangular area 101, the face image detection area The width w _i and the face image detection area height h _i are calculated, and the face image determination unit 131 calculates the xy coordinate values x _i and y _i , the face image detection area width w _i , and the face image detection area height. supplies and h _i and a predetermined score value _{S i} in subsequent determination index value calculation section 132.

  As shown in FIG. 4I, the face image determination unit 131 performs the same processing as the above-described face image determination processing on the image 100D that has been reduced until the length of one side of the image becomes the same as the length of one side of the rectangular area 101. As shown in FIG. 4J, the rectangular region 101 is made to reach the right end of the image 100D, and then the series of processes is terminated.

  Here, the processing operation of the face image determination unit 131 will be described with reference to the flowchart of FIG.

  In step S <b> 1, image data is input to the face image determination unit 131 via the image input unit 12 based on the control operation of the control unit 16.

  In step S <b> 2, the face image determination unit 131 performs a scan and cutout process on the rectangular region 101 that is a part of the image based on the image data.

In step S <b> 3, the face image determination unit 131 determines whether an image existing in the rectangular area 101 in the image is a face image by calculating a score value S _i . In step S3, when the face image determination unit 131 determines that the image existing in the rectangular area 101 is a face image, the process proceeds to step S4. On the other hand, if the face image determination unit 131 determines in step S3 that the image existing in the rectangular area 101 is not a face image, the process proceeds to step S5.

In step S4, the face image determination unit 131, xy coordinate values _x i in the rectangular region 101 the upper left corner, it calculates _{y i,} the face image detection area width _{w i,} and the face image detecting area height _{h i,} xy coordinate values x _i , y _i , face image detection region width w _i , face image detection region height h _i and score value S _i are supplied to the discrimination index value calculation unit 132.

  In step S5, the face image determination unit 131 moves the rectangular area 101 on the image by two pixels in the right direction or the downward direction.

  In step S <b> 6, the face image determination unit 131 determines whether the scan and cutout processing of the rectangular area 101 has reached the lower right corner of the image, that is, whether the scan and cutout processing of the rectangular area 101 in the entire image has ended. Judging. In step S6, if the face image determination unit 131 determines that the scanning and clipping process of the rectangular area 101 has reached the lower right end of the image, the process proceeds to step S7. On the other hand, if the face image discriminating unit 131 determines in step S6 that the scan and cut-out processing of the rectangular area 101 has not reached the lower right end of the image, the process returns to step S2, and the processing operations of steps S2 to S6 are performed. repeat.

  In step S7, the face image determination unit 131 reduces the image at a predetermined reduction rate.

  In step S8, the face image determination unit 131 determines whether one side of the image is shorter than one side of the rectangular area. In step S8, if the face image determination unit 131 determines that one side of the image is shorter than one side of the rectangular area, the process proceeds to step S9 and ends a series of processing operations. On the other hand, in step S8, when the face image determination unit 131 determines that one side of the image is not shorter than one side of the rectangular area, the process returns to step S2 and the processes of steps S2 to S8 are performed. Repeat the operation.

Note that the face image determination unit 131 described above, although the calculated score value S _i was supplied to the subsequent determination index value calculation section 132 as a determination result of the face image, based on the score value S _i, If it is a face image If it is determined, 1 of the digital data may be supplied to the subsequent determination index value calculation unit 132 if it is determined that the image is not a face image.

  The face image discriminating unit 131 detects face images having different detection positions and detection sizes, for example, as shown in FIG. 6B from the input image shown in FIG. 6A by such face image discrimination processing. be able to.

  In the face image discrimination processing described above, the face image discrimination unit 131 detects a face image a plurality of times at a position slightly shifted from the original face image area as shown in FIG. A face image having a size slightly smaller than the face image is detected.

  As described above, as a process of selecting one face image from face images detected a plurality of times for the same face image, for example, a face image having a maximum predetermined score value indicating the detection accuracy of the face image is selected. Processing (hereinafter referred to as “score value maximum region selection processing”). However, in this score value maximum area selection process, as shown in FIG. 8A, only discrete detection results based on the shift unit and the image reduction magnification in the image of the rectangular area 101 can be obtained.

  Therefore, in the face image detection apparatus 1 according to an embodiment to which the present invention is applied, a plurality of face image detection areas are averaged regardless of the discrete value based on the shift unit of the face image detection area or the image reduction magnification. Then, a process (hereinafter referred to as a face image detection area stabilization process) for obtaining a stabilized face image detection area (hereinafter referred to as a face image detection area) is applied.

  In the face image detection unit 13, the face image region averaging unit 134 performs, for example, the following calculation processing after the processing in the discrimination index value calculation unit 132 and the face image detection region classification unit 133 which will be described later. Stabilize the image detection area.

That is, the face image detection area averaging unit 134 includes the face image number i, the xy coordinate values x _i and y _i at the upper left corner of the face image detection area, the face image detection area width w _i , and the face image detection area height h. face image detecting area by using the _{i R i (x i, y} i, w i, h i) may exhibit, x coordinate value at n times the detected face image detecting area _R 1 _{~R n x} 1 ~x _n , Y coordinate values y _{1 to} y _n , face image detection area widths w _{1 to} w _n , and face image detection area heights h ₁ to h _n , a predetermined score value s calculated by the face image determination unit 131. _The average value R _ave (x _ave , y _ave , w _ave , h _ave ) is calculated by weighting with _i (formula (1) to formula (4)).

The face image detection area averaging unit 134 calculates an average value R _ave (x _ave , y _ave , w _ave , h _ave ) weighted with a predetermined score value S _i for a plurality of face image detection areas, for example, As shown in FIG. 8B, it is possible to obtain a face image area in which a plurality of face image detection areas are averaged regardless of a discrete value based on the shift unit of the face image detection area or the image reduction magnification. it can.

  For example, when the face image determination unit 131 determines that the predetermined image is a face image and outputs 1 of digital data or when the calculation process is simplified, a plurality of detection results are simply obtained. The face image detection area may be stabilized by taking the average of the above. In this case, Equations (1) to (4) are rewritten to Equations (5) to (8), respectively.

  An example of a method for selecting a face image detection area to be incorporated in calculation of face image detection area averaging from a plurality of face image detection areas in such face image detection area stabilization processing will be described with reference to FIG.

  As shown in FIG. 9A, for example, when the face image detection area of the face image 2A and the face image detection area of the face image 2B are close to each other, the face image detection area stabilization processing is performed on the face image 2A. The face image detection area 20A and the face image detection area 20B for the face image 2B must be correctly determined as to be employed in the averaging calculation as the face image detection area in any face image.

  Also, as shown in FIG. 9B, for example, when a target detection area 20C different from the face image 2A exists in the face image detection area 20A of the face image 2A, or a face image detection area of the face image 2A When 20A and the detection area 20C substantially overlap, in the face image detection area stabilization process, the detection area 20C that is different from the face image 2C is excluded from the averaging calculation as a false detection area, and the face image is detected correctly. Only the detection area 20A has to be incorporated into the calculation of the stabilization process.

  Based on such an example, in the face image detection region stabilization processing in the face image detection device 1, any two face image detection regions are incorporated in the calculation of averaging as, for example, (1) the same person's face image detection region. (2) the face image detection area of different persons is incorporated in the calculation of averaging in each person, and (3) the face image is excluded from the calculation of averaging as an erroneous detection area. Detection area averaging is calculated.

  As shown in FIG. 10, the discrimination index value calculation unit 132 arbitrarily selects, for example, a reference area 200 </ b> A serving as a discrimination reference and a discrimination target area 200 </ b> B serving as a discrimination target from among a plurality of face image detection areas. The detected area size index value b / a, which is indicated by using one side a of 200A and one side b of the discriminated area 200B and becomes an index value of the relative size of the reference area 200A and the discriminated area 200B, and the reference area 200A and the discriminated area 200B A center-to-center distance index value c / a, which is indicated by using the center-to-center distance c with the discrimination area 200B and one side a of the reference area 200A and becomes an index value of the relative distance between the reference area 200A and the discrimination area 200B, is calculated. To do.

  Then, the discrimination index value calculation unit 132 supplies the plurality of face image detection areas, the detection area size index value b / a, and the center distance index value c / a to the subsequent face image detection area classification unit 133.

  That is, the discrimination index value calculation unit 132 calculates the detection area size index value b / a and the center distance index value c / a for all combinations of the reference area 200A and the discrimination area 200B in the plurality of face image detection areas. Each time the detection area size index value b / a and the center distance index value c / a in all combinations of the reference area 200A and the determination area 200B are calculated, the detected area size index value b / a is supplied to the subsequent face image detection area classification unit 133. .

  Note that the discrimination index value calculated by the discrimination index value calculation unit 132 is not limited to these, and is, for example, the overlapping area with respect to the total area of the two face image detection regions, the Manhattan distance between the center points of the two face image detection regions, or the like. May be. Further, the number of determination index values calculated is not limited.

  FIG. 11 is a diagram showing a relative index composed of a relative size and a relative position of the discrimination area 200B with respect to the reference area 200A determined according to the detection area size index value b / a and the center distance index value c / a. is there.

  The face image detection area classifying unit 133 is set by setting, for example, boundary lines 131, 132, and 133 as shown in FIG. 12 as relative indicators of the discrimination area 200B with respect to the reference area 200A as shown in FIG. The areas 13A, 13B, 13C, and 13D are classified.

  As a method of setting each boundary line by the face image detection region classification unit 133, for example, a method of setting by statistical learning using a heuristic method may be used. As a more appropriate method, for example, sufficient For example, a classification tag generated using a large amount of face image data may be applied to SVM or LDA (Linear Discriminant analysis) as teaching data and statistically learned.

  The face image detection area classification unit 133 determines the reference area 200A and the determination area 200B having the relative index classified as the area 13A as the face image detection area of the same person.

  Further, the face image detection area classification unit 133 determines that the reference area 200A and the determination area 200B having the relative index classified as the area 13B are the face image detection areas of different persons.

  Further, the face image detection area classification unit 133 determines the determined area 200B as an erroneously detected area for the reference area 200A and the determined area 200B having the relative indexes classified into the areas 13C and 13D.

  For example, for the relative index 31, the face image detection area classification unit 133 has the same size of the reference area 200A and the size of the determination area 200B (detection area size index value b / a = 1), and the reference area Since the center-to-center distance between 200A and the determination area 200B is short (the center-to-center distance index value c / a is small), it is determined as the face image detection area of the same person.

  Further, the face image detection area classification unit 133, for example, for the relative index 32, the size of the determination area 200B is smaller than the size of the reference area 200A (detection area size index value b / a <1), and the reference area Since the center-to-center distance between the 200A and the determination area 200B is long (the center-to-center distance index value c / a is large), it is determined as a face image detection area of a different person.

  Further, the face image detection area classification unit 133, for example, for the relative index 33, the size of the determination area 200B is larger than the size of the reference area 200A (detection area size index value 1 <b / a), and the reference area Since the center-to-center distance is so long that 200A and the discrimination target area 200B almost overlap each other (the center-to-center distance index value c / a is medium), the discrimination target area 200B is discriminated as an erroneous detection area.

The face image detection area classification unit 133 supplies the relative indices of the classified reference area 200A and the determined area 200B to the subsequent face image detection area stabilization processing unit 134. The face image detection area classification unit 133 also has an xy coordinate value x _i , y _i , a face image detection area width w _i , a face image detection area height h _i, and a score value S _i at the upper left corner of the face image detection area. Alternatively, digital data (1 or 0) is supplied to the face image detection region stabilization processing unit 134.

  As described above, the face image detection area averaging unit 134 uses the formula (1) to the formula (4) or the determination area 200B having the relative indices classified by the face image detection area classification unit 133. It is possible to stabilize the detection result of the face image by calculating the averaging of the face image detection area using Expression (5) to Expression (8).

  In other words, the face image detection region stabilization processing unit 134 applies only the reference region 200A as the face image detection region of the same person to the reference region 200A and the discriminated region 200B having the relative indices classified into the region 13A. For the reference area 200A and the discriminated area 200B having the relative indices classified into the area 13B, the averaging calculation is separately performed as a face image detection area of a different person, and the area 13C is calculated. And for the reference area 200A and the discriminated area 200B having the relative index classified into the area 13D, the discriminated area 200B is deleted from the averaging calculation as an erroneous detection, and only the reference area 200A is averaged To perform the calculation.

  The face image detection area averaging unit 134 generates a face image in which the face image detection area is stabilized by performing calculation for averaging such areas, and supplies the face image to the image output unit 15.

  Here, an experimental result in which the accuracy of stabilization of the face image detection area is compared between the score value maximum area selection process and the face image detection area stabilization process is shown.

  FIG. 13 shows an intermediate point (right eye position) between the eyes and corners of the right eye in the face image detection area of 541 samples detected by Adaboost, an intermediate point (left eye position) between the eyes and corners of the left eye, and the positions of the front teeth ( It is a figure which shows the dispersion | variation in (mouth position).

  As shown in FIG. 13, the distribution of the right eye position, the left eye position, and the mouth position of the face image detection area of 541 samples subjected to the face image detection area stabilization process is obtained when the score value maximum area selection process is performed. The distribution width is smaller than the distribution of the right eye position, the left eye position, and the mouth position of the face image detection area of the 541 samples of the face image. It can be seen that the detection area is stabilized.

  Table 1 shows the results of calculating the distance from the average position (error average) and the variance (error variance value) from the average position at the right eye position, left eye position, and mouth position of the 541 samples. .

  As shown in Table 1, the error average and error variance in the right eye position, left eye position, and mouth position of the 541 samples are the values when the face image detection region stabilization processing is performed. The value is smaller than the value when applied, and it can be seen that the face image detection area is more stabilized in the face image detection area stabilization process than in the maximum score value area selection process.

  In the face image detection apparatus 1 according to an embodiment to which the present invention is applied, when a plurality of face image detection regions are detected, the average unit of the plurality of face image detection regions is calculated, thereby shifting the face image detection region shift unit. It is possible to obtain a face image area in which a plurality of face image detection areas are stabilized, regardless of discrete values based on the image reduction magnification.

  Further, in the face image detection device 1, since the face image determination unit 131 moves the rectangular area 101 in units of pixels and reduces the input image at a predetermined reduction rate, the position resolution and image reduction rate that are finer than the pixel shift amount. It is possible to obtain a stable face image area detection result with a resolution of a smaller area size shift.

  Further, the face image detection apparatus 1 has a smaller amount of calculation than the method of stabilizing the face image area by detecting the face image area after detecting the face image area. False detection can be prevented, and the face detection area can be stabilized with high accuracy.

  Further, the face image detection device 1 can improve the identification performance of the face image because it does not consider the problem of vector dimension shift that may occur when using the vector feature amount in the unique face image.

  In addition, the face image detection device 1 can limit the search range of the face part image by using a stable face image detection area as a pre-processing for detecting the face part image, and can calculate the area stabilization process. The amount can be reduced and the recognition error rate can be reduced.

  It should be noted that the present invention is not limited to the above-described embodiments, and various modifications can be made without departing from the scope of the present invention.

  For example, in the above-described embodiment, the present invention is applied to a face image detection device that detects a face image from an image based on input image data and outputs a detection result of a face image detection region or the like. The image detected from the image based on the image data is not limited to the face image, and may be any image such as a whole image of a person and a hand.

  Further, for example, in the above-described embodiment, the face image discrimination unit 131 uses the 20 × 20 pixel rectangular area 101 for discriminating the face image in the 80 × 100 pixel image 100A based on the input image data. The image 100A is moved rightward or downward by two pixels from the upper left end of the image 100A to reach the lower right end of the image 100A. The pixel size of the image based on the input image data, the image size of the rectangular area, The moving width and moving direction of the rectangular area are not limited to this.

It is a figure which shows the face image detection apparatus in one Embodiment to which this invention is applied. It is a block diagram which shows the internal structure of the face image detection apparatus in one Embodiment to which this invention is applied. It is a block diagram which shows the structure of a face image detection part. It is a figure for demonstrating the process of a face image discrimination | determination part. It is a flowchart for demonstrating the processing operation of a face image discrimination | determination part. (A) is an example of an input image, (B) is a figure which shows a mode that a face image is detected from an input image. It is a figure which shows a mode that a face image is detected in multiple times in the position slightly shifted | deviated from the original face image area | region. (A) is a figure which shows the face image detection area | region by a score value maximum area | region selection process, (B) is a figure which shows the face image detection area | region by a face image detection area stabilization process. In the face image detection area stabilization process, it is a figure which shows the example of a method of selecting the face image detection area incorporated in calculation of a face image detection area stabilization process from several face image detection areas. It is a figure for demonstrating the calculation example of a discrimination | determination index value. It is a figure which shows the relative parameter | index of two face image detection areas. It is a figure which shows a mode that the relative parameter | index of two face image detection areas is classified into each area | region set by the face image detection area classification | category part. It is a figure which shows the dispersion | variation in the part of the face in the face image detection area | region of 541 samples detected by Adaboost.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Face image detection apparatus, 11 Operation part, 12 Image input part, 13 Face image detection part, 14 Recording part, 15 Image output part, 16 Control part, 131 Face image discrimination | determination part, 132 Discrimination index value calculation part, 133 Face image Detection region classification unit, 134 Face image region averaging unit 134

Claims (7)

In an image detection device that detects a predetermined image from an input image,
Each time the relative size between the detection target area of a predetermined size and the input image is changed, the detection target area is scanned and cut out from the input image so as to move by a predetermined pixel, and the detection target area within the cut out detection target area Image discriminating means for discriminating whether an image is the predetermined image and outputting a discrimination result;
When the image determining means determines that the image in the detection target area is the predetermined image, the relative position of the two detection areas including the reference area and the determination target area detected for the predetermined image, and Index value calculating means for calculating a predetermined index value based on the relative size;
Detection for determining whether or not the determination target area classified based on the predetermined index value calculated by the index value calculation means is adopted in a calculation process for generating a detection result of the predetermined image An image detection apparatus comprising: an area determination unit.   When the relative distance between the two detection areas is short and the sizes of the two detection areas are substantially the same, the detection area determination means determines the determination area as the two detection areas as the same detection area. The image detection apparatus according to claim 1, wherein it is determined that the calculation process is adopted.   The detection area determination means determines that only the reference area is adopted for the calculation process when the relative distance between the two detection areas is short and the sizes of the two detection areas are greatly different. The image detection apparatus according to claim 1.   When the relative distance between the two detection areas is long and the sizes of the two detection areas are greatly different, the detection area determination means adopts the reference area and the determination area as separate calculation processes. The image detection apparatus according to claim 1, wherein the determination is performed. An average value of the positions and sizes of a plurality of the detection regions including the discriminated region determined to be adopted in the calculation process by the detection region determination means, and a detection region in which the positions and sizes are averaged The image detection apparatus according to claim 1, further comprising detection area averaging means for generating   The image detection apparatus according to claim 1, wherein the predetermined image is a face image. In an image detection method for detecting a predetermined image from an input image,
Each time the relative size between the detection target area of a predetermined size and the input image is changed, the detection target area is scanned and cut out from the input image so as to move by a predetermined pixel, and the detection target area within the cut out detection target area An image determination step of determining whether an image is the predetermined image and outputting a determination result;
When it is determined in the image determination step that the image in the detection target area is the predetermined image, the relative positions of the two detection areas including the reference area and the determination target area detected for the predetermined image And an index value calculating step for calculating a predetermined index value based on the relative size;
It is discriminated whether or not the discriminated area classified based on the predetermined index value calculated in the index value calculating step is adopted in a calculation process for generating a detection result of the predetermined image. An image detection method comprising: a detection region discrimination step.

Images