JP6121768B2 - Image detection apparatus, control program, and image detection method - Google Patents

Description

  The present invention relates to a technique for detecting a detection target image from a processing target image.

  Patent Documents 1 to 3 disclose techniques for detecting a detection target image from a processing target image.

JP 2008-27058 A JP 2009-175821 A JP 2011-221791 A

  Now, when detecting a detection target image from a processing target image, improvement of the detection accuracy is desired.

  Therefore, the present invention has been made in view of the above points, and an object thereof is to provide a technique capable of improving the detection accuracy of a detection target image.

  In order to solve the above-described problem, an aspect of the image detection device according to the present invention is an image detection device that detects a detection target image from a processing target image, the accuracy value indicating the probability as the detection target image. A map generation unit that generates a map indicating a distribution in the processing target image; a binarization processing unit that binarizes the map using a threshold value to generate a binarized map; and An extraction unit that extracts a circular area from the first area in the binarization map generated by using the threshold corresponding to an area having an accuracy value equal to or greater than the threshold or greater than the threshold. A first evaluation value indicating an overlapping area between the first region and the circular region extracted from the first region of the binarized map by the extraction unit, and the first region by the circular region and the extraction unit. Extracted from the region Based on a second evaluation value indicating an overlapping area with another circular region, a threshold value adjusting unit that adjusts the threshold value, and the threshold value adjusted by the threshold value adjusting unit is used. And a specifying unit that specifies the detection target image in the processing target image based on the first region of the binarized map generated in the above.

  Here, the circular area is a concept including not only a regular circular area but also an elliptical area.

  In one aspect of the image detection apparatus according to the present invention, the threshold adjustment unit overlaps the first region with a circular region extracted from the first region of the binarization map by the extraction unit. A first evaluation value indicating an area, a second evaluation value indicating an overlapping area between the circular region and another circular region extracted from the first region by the extraction unit, the circular region, and the map, A third value indicating an overlapping area with the second region in the binarization map corresponding to a region where the accuracy value is less than or less than the threshold value used in the generation of the binarization map. The threshold value is adjusted based on the evaluation value.

  In the image detection device according to the aspect of the invention, the extraction unit may detect an edge of the binarized map and perform the first region of the binarized map by a Hough transform using the coordinates of the edge. Identify the circular area at.

  An aspect of the control program according to the present invention is a control program for controlling an image detection device that detects a detection target image from a processing target image. The control program includes (a) the detection target image. A step of generating a map indicating the distribution of the accuracy value indicating the probability in the processing target image, and (b) generating a binarized map by binarizing the map using a threshold value And (c) a portion of the binarized map generated by using the threshold corresponding to an area where the accuracy value is equal to or greater than the threshold or greater than the threshold in the map. A step of extracting a circular region from the region; (d) a first evaluation value indicating an overlapping area between the circular region extracted from the partial region of the binarized map and the partial region in step (c); Yen Adjusting the threshold based on a second evaluation value indicating an overlapping area between the region and another circular region extracted from the partial region in the step (c), and (e) the step ( d) performing the step of identifying the detection target image in the processing target image based on the partial region of the binarization map generated using the threshold value adjusted in d) It is.

  An aspect of the image detection method according to the present invention is an image detection method for detecting a detection target image from a processing target image, wherein: (a) the process for the accuracy value indicating the probability as the detection target image (B) generating a binarized map by binarizing the map using a threshold value; and (c) the accuracy value in the map. Extracting a circular region from a partial region in the binarization map generated using the threshold value, corresponding to a region where is greater than or equal to the threshold value, and (d ) A first evaluation value indicating an overlap area between the circular area extracted from the partial area of the binarized map in the step (c) and the partial area, and the circular area and the part in the step (c) Other extracted from region A step of adjusting the threshold value based on a second evaluation value indicating an overlapping area with the circular region; and (e) generated using the threshold value adjusted in the step (d). Identifying the detection target image in the processing target image based on the partial region of the binarized map.

  According to the present invention, it is possible to improve detection accuracy for a detection target image.

It is a figure which shows the structure of an image detection apparatus. It is a figure which shows the structure of the several functional block with which an image detection apparatus is provided. It is a figure which shows the structure of a detection part. It is a figure for demonstrating operation | movement of a detection part. It is a figure for demonstrating operation | movement of a detection part. It is a figure for demonstrating operation | movement of a detection part. It is a figure for demonstrating operation | movement of a detection part. It is a figure for demonstrating operation | movement of a detection part. It is a figure for demonstrating operation | movement of a detection part. It is a figure which overlaps and shows a detection result frame on a process target image. It is a figure for demonstrating the production | generation method of an output value map. It is a figure for demonstrating the production | generation method of an output value map. It is a figure which shows an example of an output value map. It is a figure which shows an example of a process target image typically. It is a figure which shows an example of an output value map. It is a figure which shows an example of a binarization map. It is a figure which shows the circumscribed rectangle set to the high-accuracy area | region of a binarization map. It is a figure which shows a mode that the circumscribed rectangle in a binarization map was set to the process target image. It is a figure which shows an example of a binarization map. It is a figure which shows the circumscribed rectangle set to the high-accuracy area | region of a binarization map. It is a figure which shows a mode that the circumscribed rectangle in a binarization map was set to the process target image. It is a figure which shows an example of a binarization map. It is a figure which shows the circumscribed rectangle set to the high-accuracy area | region of a binarization map. It is a figure which shows a mode that the circumscribed rectangle in a binarization map was set to the process target image. It is a flowchart which shows operation | movement of an image detection apparatus. It is a figure which shows an example of an edge map. It is a figure which shows a mode that the circular area | region extracted from the binarization map was set to the said binarization map. It is a figure which shows a mode that the circular area | region extracted from the binarization map was set to the said binarization map.

  FIG. 1 is a diagram illustrating a configuration of an image detection apparatus 1 according to an embodiment. The image detection apparatus 1 according to the present embodiment detects a detection target image from an image indicated by input image data. The image detection apparatus 1 is used, for example, in a surveillance camera system, a digital camera system, or the like. In the present embodiment, the detection target image is, for example, a human face image. Hereinafter, simply speaking “face image” means a human face image. An image to be subjected to detection processing of a detection target image is referred to as a “processing target image”.

  The image detection apparatus 1 is a kind of computer, and includes a CPU (Central Processing Unit) 10 and a storage unit 11 as shown in FIG. The storage unit 11 is configured by a non-transitory recording medium that can be read by the image detection apparatus 1 (CPU 10), such as a ROM (Read Only Memory) and a RAM (Random Access Memory). The storage unit 11 stores a control program 12 for controlling the operation of the image detection apparatus 1 and the like. The storage unit 11 may include a computer-readable non-transitory recording medium other than the ROM and RAM. The storage unit 11 may include, for example, a hard disk drive, an SSD (Solid State Drive), a USB (Universal Serial Bus) memory, and the like. Various functions of the image detection apparatus 1 are realized by the CPU 10 executing the control program 12 in the storage unit 11. In the image detection apparatus 1, a plurality of functional blocks as shown in FIG. 2 are formed by executing the control program 12.

  As shown in FIG. 2, the image detection device 1 includes, as functional blocks, an image input unit 2, a detection unit 3, a map generation unit 4, a binarization processing unit 5, and a detection target image specifying unit 6. , A threshold adjustment unit 7 and a circular region extraction unit 8 are provided. Various functions provided in the image detection apparatus 1 may be realized by hardware circuits instead of function blocks.

  The image input unit 2 is sequentially input with a plurality of image data respectively indicating a plurality of images sequentially captured by an imaging unit (camera) included in the surveillance camera system or the like. The image input unit 2 outputs image data indicating the processing target image. The image input unit 2 may use each image obtained by the imaging unit as a processing target image, or may use an image obtained every few seconds among images obtained by the imaging unit as a processing target image. In the imaging unit, for example, L (L ≧ 2) images are captured per second. That is, the imaging frame rate at the imaging unit is Lfps (frame per second). In the image captured by the imaging unit, M (M ≧ 2) pixels are arranged in the row direction, and N (N ≧ 2) pixels are arranged in the column direction. The resolution of the image picked up by the image pickup unit is, for example, VGA (Video Graphics Array), and M = 640 and N = 480.

  In the following, in an image captured by the imaging unit, the size of an area where m pixels (m ≧ 1) are arranged in the row direction and n pixels (n ≧ 1) are arranged in the column direction is mp × np. (P means pixel). In addition, among a plurality of values arranged in a matrix, a value located in the m-th row and the n-th column with reference to the upper left may be referred to as an m × n-th value.

  The detection unit 3 uses the image data output from the image input unit 2 to detect a face image for the processing target image. The map generation unit 4 generates an output value map indicating the distribution in the processing target image with respect to the detection accuracy value indicating the likelihood as the face image based on the detection result of the detection unit 3.

  The binarization processing unit 5 binarizes the output value map generated by the map generation unit 4 using a threshold value to generate a binarized map. The circular area extraction unit 8 is a binarized map generated by using the threshold corresponding to an area where the detection accuracy value is equal to or greater than the threshold or an area greater than the threshold in the output value map. Extract a circular region from the region at.

  Here, the circle is a concept including not only a perfect circle (perfect circle) but also an ellipse. The circular region is a concept including not only a regular circular region but also an elliptical region.

  Further, in the output value map, the region in the binarization map generated by using the threshold corresponding to the region where the detection accuracy value is greater than or equal to the threshold or larger than the threshold is expressed as “ This is called “high accuracy region”. The circular area extraction unit 8 extracts a circular area from the high accuracy area of the binarized map. Further, in the output value map, the region in the binarization map generated by using the threshold corresponding to the region where the detection accuracy value is less than the threshold or the region below the threshold is expressed as “low”. This is called “accuracy region”. When the high accuracy region corresponds to a region where the detection accuracy value is greater than or equal to the threshold value in the output value map, the low accuracy region corresponds to a region where the detection accuracy value is less than the threshold value in the output value map . On the other hand, when the high accuracy region corresponds to the region where the detection accuracy value is larger than the threshold value in the output value map, the low accuracy region is detected when the detection accuracy value is equal to or less than the threshold value in the output value map. Corresponds to the region.

  The threshold adjustment unit 7 is based on the binarization map generated by the binarization processing unit 5 and the circular region extracted from the high-accuracy region of the binarization map by the circular region extraction unit 8. A threshold value used for binarization in the binarization processing unit 5 is adjusted. The detection target image specifying unit 6 uses the threshold adjusted by the threshold adjustment unit 7 to generate a detection target image in the processing target image based on the binarization map generated by the binarization processing unit 5. Is identified. As a result, the image detection apparatus 1 detects a face image from the processing target image.

  Next, the operation of each block of the image detection apparatus 1 will be described in detail.

<Detection process>
FIG. 3 is a diagram illustrating a configuration of the detection unit 3. As shown in FIG. 3, the detection unit 3 includes a feature amount extraction unit 30 and a discriminator 31. The detection unit 3 uses the detection frame to perform detection processing for detecting, as a detection result region, a region that is highly likely to be a face image having the same size as the detection frame in the processing target image. Hereinafter, simply speaking “detection process” means this detection process in the detection unit 3. The detection unit 3 uses a plurality of types of detection frames having different sizes in order to detect face images of various sizes in the processing target image. In the detection unit 3, for example, 30 types of detection frames are used. Each detection frame is, for example, a square.

  In the present embodiment, as will be described later, the feature amount extraction unit 30 extracts a feature amount from an image. The feature amount extraction unit 30 needs to use an image having a reference size (normalized size) as a target image from which feature amounts are extracted.

  On the other hand, in the present embodiment, the plurality of types of detection frames having different sizes include a detection frame having the same size as the reference size and a detection frame having a size different from the reference size. Hereinafter, a detection frame having the same size as the reference size is referred to as a “reference detection frame”, and a detection frame having a size different from the reference size is referred to as a “non-reference detection frame”. In the present embodiment, the detection frame having the smallest size among the plurality of types of detection frames is the reference detection frame. Therefore, the size of the non-reference detection frame is larger than the reference size. The size of the reference detection frame is, for example, 16p × 16p. The plurality of types of detection frames include, for example, a non-reference detection frame having a size of 18p × 18p and a non-reference detection frame having a size of 20p × 20p.

  In the present embodiment, when the detection unit 3 performs the detection process on the processing target image using the reference detection frame, the detection unit 3 moves the reference detection frame relative to the processing target image while moving the reference detection frame within the reference detection frame. A face image is detected for the image to determine whether the image is highly likely to be a face image. And the detection part 3 makes the area | region (image in a reference | standard detection frame) determined with possibility that it is a face image high in a process target image as a detection result area | region.

  On the other hand, when performing detection processing using the non-reference detection frame for the processing target image, the detection unit 3 changes the size of the non-reference detection frame so that the size matches the reference size. Then, the detection unit 3 changes the size of the processing target image in accordance with the size change of the non-reference detection frame. The detection unit 3 detects the face image for the image in the non-reference detection frame while moving the non-reference detection frame whose size has been changed with respect to the processing target image whose size has been changed, It is determined whether or not the image is likely to be a face image. Then, the detection unit 3 performs the size change based on the region (the image in the non-reference detection frame after the size change) that is determined to be highly likely to be a face image in the processing target image that has undergone the size change. A region that has a high possibility of being a face image in a processing target image of an original size that is not known is identified, and the region is set as a detection result region.

  Hereinafter, the processing target image after the size change when the non-reference detection frame is used for the processing target image and the detection processing is performed is referred to as a “size-changed image”. Further, the non-reference detection frame after the size change when the detection process is performed using the non-reference detection frame for the processing target image is referred to as a “size change detection frame”.

  Thus, in the present embodiment, the operation of the detection unit 3 when the detection unit 3 performs detection processing on the processing target image using the reference detection frame, and the detection unit 3 performs processing on the processing target image. Thus, the operation of the detection unit 3 when performing the detection process using the non-reference detection frame is different. The operation of the detection unit 3 will be described in detail below.

  In the detection unit 3, when the reference detection frame is used for the detection process, the feature amount extraction unit 30 sets a reference detection frame for the processing target image, and within the reference detection frame in the processing target image. A plurality of feature amounts are extracted from the image. On the other hand, when a non-reference detection frame is used for the detection process, the feature amount extraction unit 30 resizes the non-reference detection frame with respect to a size-changed image obtained by resizing the processing target image. The size change detection frame obtained in this way is set, and a plurality of feature amounts are extracted from the image in the size change detection frame in the size change image. Hereinafter, the image in the reference detection frame and the image in the size change detection frame from which the feature amount is extracted may be collectively referred to as “in-frame image”.

  Here, since the size of the reference detection frame matches the reference size, the size of the image in the reference detection frame in the processing target image becomes the reference size. In addition, since the size of the size change detection frame matches the reference size, the size of the image in the size change detection frame in the size change image becomes the reference size. Therefore, the feature amount extraction unit 30 can always extract the feature amount from the image of the reference size. The feature quantity extraction unit 30 extracts feature quantities such as Haar-like feature quantities and LBP (Local Binary Pattern) feature quantities from the in-frame image.

  The discriminator 31 includes a feature vector composed of a plurality of feature amounts extracted from the in-frame image by the feature amount extraction unit 30, and a weight vector composed of a plurality of weight coefficients generated based on the learning sample (learning sample image). Based on the above, a detection accuracy value indicating the likelihood that the in-frame image is a face image is calculated. Specifically, the feature quantity extraction unit 30 obtains an inner product of the feature vector and the weight vector for the in-frame image, and adds a real value obtained by adding a predetermined bias value to the in-frame image. Is a detection accuracy value indicating the probability that is a face image. The detection accuracy value calculated by the discriminator 31 indicates the face image likelihood (face likelihood) of the image within the reference detection frame or the image within the size change detection frame. For the discriminator 31, for example, SVM (Support Vector Machine) or Adaboost is used.

  If the calculated detection accuracy value is greater than or equal to the threshold value, the classifier 31 determines that the possibility that the in-frame image is a face image is high. That is, when the reference detection frame is used, the discriminator 31 determines that the image in the reference detection frame in the processing target image is a region that is highly likely to be a face image having the same size as the reference detection frame. To do. Further, when the non-reference detection frame is used, the discriminator 31 is an area in which the image in the size change detection frame in the size change image is likely to be a face image having the same size as the size change detection frame. Judge that there is.

  On the other hand, the discriminator 31 determines that there is a high possibility that the in-frame image is not a face image if the calculated detection accuracy value is less than the threshold. That is, when the reference detection frame is used, the discriminator 31 determines that the image in the reference detection frame in the processing target image is not a region that is highly likely to be a face image having the same size as the reference detection frame. To do. When the non-reference detection frame is used, the discriminator 31 determines that the image in the size change detection frame in the size change image is likely to be a face image having the same size as the size change detection frame. Judge that there is no.

  When the discriminator 31 determines that the image in the reference detection frame in the processing target image is a region that is highly likely to be a face image having the same size as the reference detection frame, the image is used as a detection result region, and the reference detection is performed. Let the frame be the detection result frame.

  If the discriminator 31 determines that the image in the size change detection frame in the size change image is an area that is highly likely to be a face image having the same size as the size change detection frame, the identifier 31 temporarily detects the outer frame of the area. The result frame. Then, the discriminator 31 specifies an area that is highly likely to be a face image having the same size as the non-reference detection frame in the processing target image that is the original image of the resized image based on the temporary detection result frame, The area is set as a detection result area, and the outer frame of the detection result area is set as a final detection result frame.

<Detection process using reference detection frame>
Next, the detection unit 3 moves the reference detection frame with respect to the processing target image, and determines whether the image in the reference detection frame is likely to be a face image. The operation will be described. 4-7 is a figure for demonstrating the said operation | movement of the detection part 3. FIG. The detection unit 3 detects a face image with respect to an image in the reference detection frame while raster scanning the reference detection frame.

  As shown in FIG. 4, the feature amount extraction unit 30 first sets a reference detection frame 100 at the upper left of the processing target image 20, and extracts a plurality of feature amounts from the image in the reference detection frame 100. The discriminator 31 obtains a detection accuracy value for an image in the reference detection frame 100 based on a feature vector composed of a plurality of feature amounts extracted by the feature amount extraction unit 30 and a weight vector composed of a plurality of weight coefficients. . Then, when the calculated detection accuracy value is equal to or greater than the threshold value, the classifier 31 determines that there is a high possibility that the region in the upper left reference detection frame 100 in the processing target image 20 is a face image. The region is set as a detection result region, and the reference detection frame 100 which is the outer frame of the region is set as a detection result frame.

  Next, the feature amount extraction unit 30 moves the reference detection frame 100 slightly to the right in the processing target image 20. For example, the feature amount extraction unit 30 moves the reference detection frame 100 to the right by one pixel or several pixels. Then, the feature quantity extraction unit 30 extracts a plurality of feature quantities from the image in the reference detection frame 100 after movement in the processing target image 20.

  After that, the discriminator 31 uses the feature vector composed of a plurality of feature amounts extracted by the feature amount extraction unit 30 and the weight vector composed of a plurality of weight coefficients for the image in the reference detection frame 100 after movement. The detection accuracy value is obtained. When the calculated detection accuracy value is equal to or greater than the threshold value, the discriminator 31 determines that the image in the reference detection frame 100 after the movement is likely to be a face image, and determines the image as the image. In addition to the detection result area, the reference detection frame 100 after movement, which is the outer frame of the image, is used as the detection result frame.

  Thereafter, the detection unit 3 operates in the same manner, and when the reference detection frame 100 moves to the right end of the processing target image 20 as illustrated in FIG. 5, the detection unit 3 detects the image in the reference detection frame 100 at the right end. The detection accuracy value is obtained. If the obtained detection accuracy value is equal to or greater than the threshold value, the detection unit 3 sets the image in the rightmost reference detection frame 100 as a detection result region, and uses the rightmost reference detection frame 100 as a detection result frame. To do.

  Next, as illustrated in FIG. 6, the feature amount extraction unit 30 moves the reference detection frame 100 to the left end of the processing target image 20 while slightly lowering the reference detection frame 100, and then extracts a plurality of features from the image in the reference detection frame 100. Extract the amount. The feature amount extraction unit 30 moves the reference detection frame 100 downward by, for example, one pixel or several pixels in the vertical direction (column direction). Thereafter, the discriminator 31 uses the feature vector composed of a plurality of feature amounts extracted from the feature amount extraction unit 30 and the weight vector composed of a plurality of weight coefficients to calculate the image in the current reference detection frame 100. Find and output the detection accuracy value. When the calculated detection accuracy value is equal to or greater than the threshold value, the discriminator 31 determines that the image in the current reference detection frame 100 is likely to be a face image, and detects the image. In addition to the result area, the reference detection frame 100 is set as a detection result frame.

  Thereafter, the detection unit 3 operates in the same manner, and when the reference detection frame 100 moves to the lower right of the processing target image 20 as illustrated in FIG. 7, the detection unit 3 moves within the lower right reference detection frame 100. The detection accuracy value for the image is obtained. If the obtained detection accuracy value is equal to or greater than the threshold value, the detection unit 3 sets the image in the lower right reference detection frame 100 as a detection result region and uses the lower right reference detection frame as the detection result frame. And

  As described above, the detection unit 3 uses the reference detection frame to detect, in the processing target image, an area that is highly likely to be a face image having the same size as the reference detection frame as a detection result area. In other words, the detection unit 3 specifies a face image having the same size as the reference detection frame in the processing target image using the reference detection frame.

<Detection process using non-reference detection frame>
When the detection unit 3 performs the detection process using the non-reference detection frame, the feature amount extraction unit 30 is configured so that the size of the non-reference detection frame matches the reference size (the size of the reference detection frame). The non-reference detection frame is resized. Then, the feature amount extraction unit 30 resizes the processing target image by the same size change ratio as the non-reference detection frame.

  In the present embodiment, since the reference size is 16p × 16p, for example, when a non-reference detection frame having a size of Rp × Rp (R> 16) is used, the feature amount extraction unit 30 performs the non-reference detection The non-reference detection frame is reduced by multiplying the vertical width (vertical width) and horizontal width (horizontal width) of the reference detection frame by (16 / R), respectively, to generate a size change detection frame. Then, the feature amount extraction unit 30 reduces the processing target image by multiplying the vertical width (number of pixels) and the horizontal width (number of pixels) of the processing target image by (16 / R), and generates a size-changed image. After that, the detection unit 3 moves the size change detection frame with respect to the size change image and moves the feature amount from the image in the size change detection frame in the same manner as the processing described with reference to FIGS. Based on the extracted feature amount, it is determined whether there is a high possibility that the image in the size change detection frame is a face image having the same size as the size change detection frame. That is, the detection unit 3 performs processing for detecting an area that is highly likely to be a face image having the same size as the size change detection frame in the size change image using the size change detection frame. Hereinafter, this process is referred to as a “size-changed version detection process”.

  In the size change version detection process, the detection unit 3 sets a size change detection frame for the size change image, and the image in the size change detection frame may be a face image having the same size as the size change detection frame. If it is determined that the property is high, the size change detection frame that is the outer frame of the image is set as a temporary detection result frame.

  In the detection unit 3, when at least one temporary detection result frame is obtained for the resized image, the discriminator 31 selects the at least one temporary detection result frame according to the processing target image that is the basis of the resized image. Convert to detection result frame.

  Specifically, the classifier 31 first sets at least one temporary detection result frame obtained for the resized image. FIG. 8 is a diagram illustrating a state in which the temporary detection result frame 130 is set for the size-changed image 120. In the example of FIG. 8, a plurality of temporary detection result frames 130 are set for the resized image 120.

  Next, as shown in FIG. 9, the classifier 31 processes the resized image 120 by enlarging (resizing) the resized image 120 in which the temporary detection result frame 130 is set and returning it to the original size. The target image 20 is converted. As a result, the temporary detection result frame 130 set in the size-changed image 120 is also enlarged, and the temporary detection result frame 130 is converted into a detection result frame 150 corresponding to the processing target image 20 as shown in FIG. The A region in the detection result frame 150 in the processing target image 20 is a detection result region that is highly likely to be a face image having the same size as the non-reference detection frame in the processing target image 20. Thereby, in the detection unit 3, based on the temporary detection result frame 130 obtained by the size-changed version detection process, a detection result region that is highly likely to be a face image having the same size as the non-reference detection frame in the processing target image. Identified.

  As described above, when the detection unit 3 performs the detection process on the processing target image using the non-reference detection frame, the non-reference detection frame that has been resized so that the size matches the reference size, The size-changed version detection process is performed using the processing target image whose size has been changed according to the size change of the reference detection frame. Thereby, even when a detection frame having a size different from the reference size is used, the feature amount extraction unit 30 can extract the feature amount from the image of the reference size. And the detection part 3 pinpoints the detection result area | region with high possibility that it is a face image of the same size as a non-reference | standard detection frame in a process target image based on the result of a size-change version detection process. Thus, the detection unit 3 performs detection processing using the non-reference detection frame.

  The detection unit 3 performs the above detection process using each of a plurality of types of detection frames. Thereby, when a face image is included in the processing target image, a detection result area (area that is highly likely to be a face image) and a detection result frame (outer frame of an area that is likely to be a face image) Is obtained, and a detection accuracy value corresponding to the detection result frame is obtained. The detection accuracy value corresponding to the detection result frame obtained for the processing target image indicates the probability that the image in the detection result frame in the processing target image is a face image.

  FIG. 10 is a diagram illustrating a state in which the detection result frame 150 obtained for the processing target image 20 is arranged so as to overlap the processing target image 20. As shown in FIG. 10, detection processing frames 150 of various sizes are obtained by performing detection processing using a plurality of types of detection frames having different sizes. This means that face images of various sizes included in the processing target image 20 are detected.

<Output value map generation processing>
The map generation unit 4 generates an output value map indicating the distribution in the processing target image with respect to the detection accuracy value indicating the likelihood (face image likelihood) as the face image based on the detection result of the detection unit 3.

  Specifically, the map generation unit 4 is a map composed of a total of (M × N) values in which M values are arranged in the row direction and N values are arranged in the column direction, similarly to the processing target image. Think about 200. Then, the map generation unit 4 sets one detection result frame for the processing target image as the target detection result frame, and sets a frame 210 having the same size as the target detection result frame to the map 200 at the same position as the target detection result frame. To set. FIG. 11 is a diagram illustrating a state in which a frame 210 is set for the map 200.

  Next, the map generation unit 4 sets “0” for each value outside the frame 210 in the map 200, and for each value within the frame 210, a detection accuracy value corresponding to the target detection result frame (the target detection result frame and The detection accuracy value obtained as a result of detecting the face image with respect to the image within the detection frame is determined. If the size of the target detection result frame is, for example, 16p × 16p, there are 16 values in the row 210, 16 in the column direction and 16 in the column direction, for a total of 256 values. Further, assuming that the size of the target detection result frame is 20p × 20p, for example, there are 20 values in the row 210 and 20 in the column direction, for a total of 400 values. FIG. 12 is a diagram for explaining a method for determining each value in the frame 210.

  The map generation unit 4 sets the value of the center 211 in the frame 210 as the detection accuracy value corresponding to the target detection result frame obtained by the detection unit 3. Then, the map generation unit 4 gradually increases the values of the other values in the frame 210 from the center 211 in the frame 210 to the outside according to a normal distribution curve with the value at the center 211 of the frame 210 as the maximum value. To be smaller. Thereby, each of a plurality of values constituting the map 200 is determined, and the map 200 corresponding to the target detection result frame is completed.

  As described above, the map generation unit 4 generates a plurality of maps 200 respectively corresponding to a plurality of detection result frames for the processing target image. And the map production | generation part 4 synthesize | combines the produced | generated several map 200, and produces | generates an output value map.

  Specifically, the map generation unit 4 adds the m × n-th values of the plurality of generated maps 200, and uses the obtained addition value as the m × n-th detection accuracy value of the output value map. . In this way, the map generation unit 4 obtains each detection accuracy value constituting the output value map. As a result, an output value map indicating the distribution of detection accuracy values in the processing target image is completed. In the output value map, similarly to the processing target image, M detection accuracy values are arranged in the row direction and N detection accuracy values are arranged in the column direction. The output value map is composed of (M × N) detection accuracy values. By referring to the output value map, it is possible to specify a region having a high likelihood of a face image in the processing target image. That is, the face image in the processing target image can be specified by referring to the output value map.

  FIG. 13 is a diagram showing an output value map for the processing target image 20 superimposed on the processing target image 20. In FIG. 13, for easy understanding, the output value map is shown by dividing the magnitude of the detection accuracy value into, for example, five stages from the first stage to the fifth stage. In the output value map shown in FIG. 13 and FIG. 15 described later, vertical line hatching is shown for the region belonging to the fifth stage where the detection accuracy value is the largest, and belongs to the second largest fourth stage. The area is shown as sand hatching. In the output value maps shown in FIG. 13 and FIG. 15, the region belonging to the third stage where the detection accuracy value is the third largest shows a right-up hatching, and the fourth largest second stage. For areas belonging to, left-upward hatching is shown. In the output value maps shown in FIGS. 13 and 15, hatching is not shown for the region belonging to the first stage having the smallest detection accuracy value.

  In the output value map shown in FIG. 13, the detection accuracy value in the region corresponding to the face image in the processing target image 20 (region in the same position as the face image) is high. This means that the face image included in the processing target image 20 is properly detected. In the output value map, in the region corresponding to the face image in the processing target image 20, the detection accuracy value at the same position as the vicinity of the center of the face image is the largest, and the detection accuracy value decreases toward the outside. It has become.

<Binarization processing>
The binarization processing unit 5 binarizes the output value map generated by the map generation unit 4 using a threshold value to generate a binarized map. Specifically, the binarization processing unit 5 changes each value in a region where the detection accuracy value is equal to or larger than the threshold value or larger than the threshold value to, for example, “1” in the output value map, and the detection accuracy value is Each value in the area less than or less than the threshold value is changed to, for example, “0”. Thus, a high accuracy region where each value is “1” corresponding to a region where the detection accuracy value is greater than or equal to the threshold value or larger than the threshold value in the output value map, and the detection accuracy value in the output value map A binarized map composed of low-accuracy regions each having a value of “0” corresponding to a region of less than or less than the threshold value is generated.

  FIG. 14 is a diagram schematically illustrating an example of the processing target image 20. FIG. 15 is a diagram showing an output value map 40 for the processing target image 20 shown in FIG. FIG. 16 is a diagram showing a binarization map 50 generated by binarizing the output value map 40 shown in FIG. 15 using a predetermined threshold value.

  As shown in FIG. 15, in the output value map 40, the detection accuracy value in the area 40 a corresponding to the face image 20 a included in the processing target image 20 and the area corresponding to the face image 20 b included in the processing target image 20. The detection accuracy value at 40b is large. On the other hand, in the output value map 40, the detection accuracy value in the region 40c corresponding to the face image 20c included in the processing target image 20 is small.

  For example, when the output value map 40 shown in FIG. 15 is binarized using, as a threshold value, a boundary value between the second stage (upward hatching) and the third stage (upward hatching) of the detection accuracy value. A binarized map 50 shown in FIG. 16 is obtained. In FIG. 16, hatching is shown in the high accuracy region 51, and hatching is not shown in the low accuracy region 52. In the output value map 40, the detection accuracy value in the region 40c corresponding to the face image 20c is generally smaller than the detection accuracy value in the regions 40a and 40b corresponding to the face images 20a and 20b. Thus, in the high accuracy region 51 of the binarized map 50, the region 51c corresponding to the face image 20c is smaller than the regions 51a and 51b corresponding to the face images 20a and 20b, respectively.

  When the threshold value used in generating the binarized map 50 is appropriately adjusted, a plurality of faces included in the processing target image 20 are included in the high accuracy area 51 of the binarized map 50 as shown in FIG. A plurality of independent (separated) regions 51a to 51c corresponding to the images 20a to 20c are included. Therefore, each of the plurality of face images 20a to 20c included in the processing target image 20 can be individually specified from the plurality of regions 51a to 51c. The threshold value used in generating the binarized map 50 is appropriately adjusted by the threshold value adjusting unit 7 as will be described later.

<Detection target image specifying process>
The detection target image specifying unit 6 uses the threshold adjusted by the threshold adjustment unit 7 to generate a processing target image based on the high accuracy region of the binarization map generated by the binarization processing unit 5. A face image is specified at. Hereinafter, the binarization map generated by using the threshold adjusted by the threshold adjustment unit 7 is particularly referred to as “specific binarization map”.

  In the present embodiment, the detection target image specifying unit 6 specifies each independent region (island region) included in the high accuracy region of the specifying binarization map. In the example of FIG. 16, each of the areas 51a to 51c is specified as an independent area. Then, the detection target image specifying unit 6 obtains a circumscribed rectangle circumscribing the independent area for each specified independent area. Each independent region included in the high-accuracy region of the specifying binarization map can be specified by performing labeling using 4-connection or the like on the specifying binarization map.

  In FIG. 17, the detection target image specifying unit 6 uses, for example, the binarization map 50 shown in FIG. 16 as the specifying binarization map, and the independent region included in the high accuracy region 51 of the binarization map 50. It is a figure which shows the said circumscribed rectangle at the time of calculating | requiring the circumscribed rectangle about 51a-51c. The circumscribed rectangles 300a to 300c shown in FIG. 17 are circumscribed rectangles of the independent areas 51a to 51c included in the high accuracy area 51 of the binarized map 50 shown in FIG.

  When the detection target image specifying unit 6 obtains a circumscribed rectangle for each independent region of the high-accuracy region of the specifying binarization map, the detection target image specifying unit 6 sets the circumscribed rectangle as a processing target image. 18 is a diagram illustrating a state in which the circumscribed rectangles 300a to 300c illustrated in FIG. 17 are set in the processing target image 20 illustrated in FIG. The detection target image specifying unit 6 determines that for each circumscribed rectangle set as the processing target image, the image in the circumscribed rectangle is one face image. Thereby, a face image is specified in the processing target image 20.

  When the processing target image is displayed on the display device, the image detection device 1 displays the circumscribed rectangle obtained by the detection target image specifying unit 6 so as to overlap the processing target image, as shown in FIG.

  Further, the image detection apparatus 1 may compare a face image registered in advance with a face image specified in the processing target image (an image in a circumscribed rectangle) and determine whether or not they match. If the face image registered in advance and the face image specified in the processing target image do not match, the image detection apparatus 1 performs mosaic processing on the face image in the processing target image. The processing target image may be displayed on the display device. As a result, when the image detection apparatus 1 according to the present embodiment is used in a surveillance camera system, even when a face image of a neighbor's person is photographed by the surveillance camera, the face image cannot be recognized. can do. That is, a privacy mask can be realized.

<Threshold adjustment processing>
If the threshold used when the binarization processing unit 5 binarizes the output value map is not appropriately set, the image detection apparatus 1 may not be able to correctly detect the face image from the processing target image. This point will be described below.

  FIG. 19 shows a binarization obtained by binarizing the output value map 40 shown in FIG. 15 with a threshold smaller than the threshold used in the generation of the binarization map 50 shown in FIG. It is a figure which shows the map.

  When the threshold value when the output value map 40 is binarized is small, a region where the detection accuracy value is not so large in the output value map 40 also becomes the high accuracy region 51. Accordingly, as shown in FIG. 19, in the high accuracy region 51, the regions 51a and 51b corresponding to the face images 20a and 20b that are close to each other may be connected to form one independent region. In this case, when a circumscribed rectangle for each independent area included in the high-accuracy area 51 of the binarization map 50 shown in FIG. 19 is obtained, an independent area composed of areas 51a and 51b is obtained as shown in FIG. A circumscribed rectangle 300d circumscribing the region and a circumscribed rectangle 300c circumscribing the region 51c are generated.

  When the circumscribed rectangles 300c and 300d are set as the processing target image 20, as shown in FIG. 21, one circumscribed rectangle 300d is set for the two face images 20a and 20b, and one for the face image 20c. Two circumscribed rectangles 300c are set. Since the detection target image specifying unit 6 sets the image within one circumscribed rectangle in the processing target image as one face image, the face image 20c can be appropriately detected from the processing target image 20, but the face The images 20a and 20b are specified as one face image, and it becomes difficult to individually detect the face images 20a and 20b.

  FIG. 22 shows a binarization obtained by binarizing the output value map 40 shown in FIG. 15 with a threshold value larger than the threshold value used in the generation of the binarization map 50 shown in FIG. It is a figure which shows the map.

  When the threshold value when the output value map 40 is binarized is large, a region where the detection accuracy value is not so large in the output value map 40 does not become the high accuracy region 51. Therefore, as shown in FIG. 22, for the face image 20 c having a small detection accuracy value in the corresponding region in the output value map 40, the region corresponding to the face image 20 c is not included in the high accuracy region 51. There is. In this case, when a circumscribed rectangle for each independent area included in the high accuracy area 51 of the binarized map 50 shown in FIG. 21 is obtained, a circumscribed rectangle circumscribing the area 51a is obtained as shown in FIG. 300a and a circumscribed rectangle 300b circumscribing the region 51b are generated.

  When the circumscribed rectangles 300a and 300b are set as the processing target image 20, the circumscribed rectangles 300a and 300b are set for the face images 20a and 20b, respectively, as shown in FIG. The rectangle is not set. Therefore, although it is possible to detect the face images 20a and 20b, it is difficult to detect the face image 20c.

  As described above, when the threshold value used for generating the binarized map is small, it is difficult to appropriately detect a plurality of face images at close distances.

  On the other hand, when the threshold value used in generating the binarized map is large, it is difficult to appropriately detect a face image having a small detection accuracy value of the corresponding region in the output value map.

  Therefore, in the present embodiment, the detection target image specifying unit 6 can specify a face image having a small detection accuracy value of the corresponding region in the output value map in the processing target image, and a plurality of short distances. The threshold value adjusting unit 7 appropriately adjusts the threshold value used in generating the binarized map so that each of the face images can be specified individually. Hereinafter, the operation of the image detection apparatus 1 when the threshold adjustment unit 7 adjusts the threshold will be described in detail.

  FIG. 25 is a flowchart showing threshold adjustment processing in the image detection apparatus 1. The threshold value adjustment process shown in FIG. 25 is executed using the output value map when the map generation unit 4 generates an output value map for the processing target image. In the threshold value adjustment process according to the present embodiment, the image detection apparatus 1 changes the threshold value in a plurality of stages to generate a binarized map at each threshold value. Then, the image detection apparatus 1 determines an appropriate threshold value to be finally used based on the plurality of generated binarization maps. In the present embodiment, the threshold value is changed, for example, by 5 to 10 steps. Therefore, in the threshold adjustment process, 5 to 10 binarized maps having different used threshold values are generated.

  In the threshold value adjustment process, as shown in FIG. 25, first, in step s1, the threshold value adjustment unit 7 temporarily sets the threshold value used in the binarization map generation in the binarization processing unit 5. To do. Here, for example, the threshold value is set to the minimum value in the change range.

  Next, in step s2, the binarization processing unit 5 binarizes the output value map generated by the map generation unit 4 using the threshold value temporarily set in step s1, and generates a binarized map. .

  Next, in step s3, the circular area extraction unit 8 extracts a circular area from the high-accuracy area of the binarized map generated in step s2. A method for extracting the circular area will be described in detail later.

  Next, in step s4, the threshold adjustment unit 7 determines the threshold temporarily set in step s1 based on the binarized map generated in step s2 and the circular area extracted in step s3. The evaluation value for determination is calculated. The evaluation value for determination is a value indicating the appropriateness of the threshold value. A method for calculating the evaluation value for determination will be described in detail later.

  Next, in step s5, the threshold adjustment unit 7 determines whether the threshold has been changed by a predetermined range (a plurality of predetermined stages). If it is determined in step s5 that the threshold value has not been changed within the predetermined range, the above-described step s1 is executed, and a new threshold value is temporarily set in the binarization processing unit 5. Here, a threshold value increased by one step is temporarily set. Thereafter, Steps s2 to s4 are executed, and an evaluation value for determination is calculated for the threshold value increased by one step. Thereafter, the image detection apparatus 1 operates in the same manner.

  When it is determined in step s5 that the threshold value has been changed by a predetermined range, in step s6, the threshold value adjusting unit 7 corresponds to each of the threshold values calculated in the threshold value adjusting process. Based on the plurality of determination evaluation values, an appropriate threshold value is determined from the plurality of threshold values. Specifically, the threshold value adjusting unit 7 determines a threshold value corresponding to the maximum value of the plurality of evaluation values for determination among the plurality of threshold values as an appropriate threshold value. As a result, the threshold value adjustment process ends.

  When an appropriate threshold value is determined in step s6, the binarization processing unit 5 binarizes the output value map using the appropriate threshold value, that is, the threshold value adjusted by the threshold value adjustment unit 7. To generate a binarization map for identification. Then, the detection target image specifying unit 6 specifies a face image in the processing target image based on the high accuracy region of the specifying binarization map generated by the binarization processing unit 5 as described above.

<About circular area extraction processing>
Next, the circular area extraction process in step s3 will be described in detail. In the present embodiment, the circular area extraction unit 8 detects the edge of the binarized map generated in step s2 (the boundary between the high accuracy area and the low accuracy area), and the Hough transform using the detected edge coordinates. To extract a circular region, for example, a regular circular region, from the high-accuracy region of the binarized map.

  In step s3, the circular area extraction unit 8 first detects an edge of the binarized map and generates an edge map indicating the edge. FIG. 26 is a diagram showing an edge map 60 showing the edges 510 of the binarized map 50 shown in FIG. The edge of the binarized map can be detected using, for example, the Canny method. Hereinafter, simply speaking an edge means an edge of a binarized map.

  In the edge map, M values are arranged in the row direction and N values are arranged in the column direction, as in the processing target image, the output value map, and the binarization map. Therefore, the edge map is composed of a total of (M × N) values. In the edge map, each value indicating an edge, that is, each value at the same position as the edge position in the binarized map is, for example, “1”, and each other value is, for example, “0”. Yes.

  When the circular area extraction unit 8 generates an edge map, the circular area extraction unit 8 obtains coordinates of each value indicating an edge in the edge map. Since the coordinates of each value indicating an edge in the edge map are the coordinates of each value constituting the edge in the binarized map, the circular area extracting unit 8 can determine whether the value in the binarized map is based on the edge map. The coordinates of each value constituting the edge are obtained.

  In this embodiment, the edge map and the binarized map are defined as an xy plane in which the upper left corner of the edge map and the binarized map is the origin, the row direction is the x-axis direction, and the column direction is the y-axis direction. ing. Then, the circular area extraction unit 8 obtains xy coordinates on the xy plane for each value indicating an edge in the edge map. Thereby, the xy coordinate on the xy plane is calculated | required about each value which comprises an edge in a binarization map. Hereinafter, the xy coordinates are referred to as “edge coordinates”.

  Next, the circular area extraction unit 8 specifies a circular area, for example, a regular circular area, in the high-accuracy area of the binarized map by Hough transform using a plurality of edge coordinates obtained for the edge. Hereinafter, a method for identifying a regular circular area using the Hough transform will be described. The edge coordinates to be described are called target edge coordinates.

  A regular circular region can be expressed by three parameters: a center x coordinate Cx, a center y coordinate Cy, and a radius r. In the Hough transform, a Hough space represented by a three-dimensional axis indicating each of these three parameters is used. Hereinafter, the three parameters expressing the regular circular area may be collectively referred to as a “circle expression parameter group”.

  The circular area extraction unit 8 performs a voting process for each of the obtained plurality of edge coordinates. In the voting process for the target edge coordinates, the circular area extraction unit 8 first selects a plurality of different types of values such that the values of the target edge coordinates are located on the circumference on the xy plane defined in the binarization map. Consider a circular area. Then, the circular area extraction unit 8 votes for three-dimensional coordinates in the Hough space indicating three parameters (a circle expression parameter group) that represent the regular circular area for each of a plurality of types of regular circular areas. I do.

  The circular area extraction unit 8 uses the following formulas for each of a plurality of different types of regular circular areas where the values of the target edge coordinates are located on the circumference in the xy plane defined in the binarization map: 1) is used to obtain a circle expression parameter group expressing the regular circular area.

r ² = (x−Cx) ² + (y−Cy) ² (1)
Here, x and y in the formula (1) indicate the x coordinate and the y coordinate of the target edge coordinates, respectively. The circular area extracting unit 8 obtains r corresponding to each set of Cx and Cy by changing each of Cx and Cy in Equation (1) in a plurality of ways. Thereby, a plurality of sets of Cx, Cy, and r are obtained. Since one set of Cx, Cy, and r is a circle expression parameter group that represents one regular circular region in which the value of the target edge coordinate is located on the circumference, there are a plurality of sets of Cx, Cy, and r. By obtaining the set, a circle expression parameter group is obtained for each of a plurality of different types of regular circular regions where the values of the target edge coordinates are located on the circumference. For example, if Cx and Cy are changed in 100 ways, for example, 10,000 pairs of Cx, Cy, and r are obtained. A circle expression parameter group for each of the circular regions is obtained. For each of the obtained plurality of circle expression parameter groups (for example, 10,000 circle expression parameter groups), the circular area extraction unit 8 performs three-dimensional coordinates in the Hough space indicating the three parameters constituting the circle expression parameter group. Vote for.

  The circular area extraction unit 8 performs such voting processing for each of the obtained plurality of edge coordinates (coordinates of a plurality of values constituting an edge in the binarized map). Then, the circular area extraction unit 8 converts the circular area represented by the three parameters constituting the three-dimensional coordinates having the largest number of votes in the Hough space into the circular area included in the high accuracy area of the binarized map. And In this way, one regular circular area is extracted from the high accuracy area of the binarized map.

  When the circular region extraction unit 8 specifies one regular circular region in the high-accuracy region of the binarized map, a plurality of edges obtained from the edge coordinates of values located on the circumference of the regular circular region are obtained. Delete from the coordinates, and perform voting again for each of the remaining edge coordinates. Then, the circular area extraction unit 8 converts the circular area represented by the three parameters constituting the three-dimensional coordinates having the largest number of votes in the Hough space into the circular area included in the high accuracy area of the binarized map. This is an area.

  Thereafter, the circular area extraction unit 8 operates in the same manner, and ends the circular area extraction process when the number of remaining edge coordinates is equal to or less than a predetermined threshold value.

  In this way, in step s3, the circular area extraction unit 8 extracts a circular area from the high-accuracy area of the binarized map generated using the threshold value temporarily set in step s1. FIG. 27 is a diagram showing the circular area 400 extracted from the binarization map 50 shown in FIG. In the example of FIG. 27, three circular regions 400 are extracted from the binarized map 50.

  The circular area extraction unit 8 may extract other circular areas such as an elliptical area instead of a regular circular area from the binarized map. An elliptical region can be extracted using the Hough transform.

<Evaluation value calculation method for determination>
Since the human face has a circular shape, the contour of the face image included in the processing target image also has a circular shape. Therefore, by appropriately adjusting the threshold value used in the generation of the binarization map, the high accuracy region of the binarization map is included in the processing target image as shown in FIG. There is a high possibility that an independent circular area corresponding to each face image is included. Therefore, in this case, the detection target image specifying unit 6 can appropriately specify each face image included in the processing target image based on the high accuracy region of the binarized map.

  On the other hand, if the threshold value used in generating the binarized map is too small, the circular area corresponding to the face image is large in the high accuracy area of the binarized map as shown in FIG. Therefore, there is a possibility that a plurality of circular areas respectively corresponding to a plurality of face images close to each other are connected, and the high accuracy area of the binarized map includes one independent area corresponding to the plurality of face images. Get higher. Therefore, in this case, it becomes difficult for the detection target image specifying unit 6 to appropriately specify each face image included in the processing target image based on the high accuracy region of the binarized map.

  Also, if the threshold value used in the generation of the binarized map is too large, the circular area corresponding to the face image becomes small in the high accuracy area of the binarized map as shown in FIG. There is a high possibility that a circular area corresponding to a face image (face image 20c in FIG. 14) having a small detection accuracy value of the corresponding area in the output value map will disappear. Therefore, in this case as well, it is difficult for the detection target image specifying unit 6 to appropriately specify each face image included in the processing target image based on the high accuracy region of the binarized map.

  As described above, if the threshold value used in the generation of the binarized map is too large, in the high-accuracy region of the binarized map, there is a high possibility that the circular region corresponding to the face image becomes small and disappears. Become. Therefore, in order to reduce this possibility, it is desirable to adjust the threshold value so that the high-accuracy region of the binarization map includes as large a circular region as possible.

  On the other hand, if the threshold value used in the generation of the binarized map is too small, a plurality of circular areas respectively corresponding to a plurality of face images are connected in the high-accuracy area of the binarized map. There is a high possibility that one independent area corresponding to the face image is included. Therefore, in order to reduce this possibility, it is desirable to adjust the threshold value so that the high-accuracy region of the binarization map includes as many independent circular regions as possible.

  Therefore, in the present embodiment, the threshold adjustment unit 7 adjusts the threshold used in generating the binarized map based on the evaluation value for determination, thereby increasing the accuracy of the binarized map. The region includes as large a circular region as possible, but includes as many independent circular regions as possible. As a result, each face image included in the processing target image can be appropriately specified individually. Hereinafter, a method for calculating the evaluation value for determination will be described in detail.

  In the present embodiment, the threshold adjustment unit 7 obtains the integrated evaluation value B using the following formula (2) for each circular region extracted in step s3.

B = A1-A2-A3 (2)
Here, in Expression (2), A1 is a target circular area for which the integrated evaluation value B is calculated using Expression (2) (hereinafter referred to as “target circular area”) and the binary value generated in Step s2. It is a 1st evaluation value which shows the overlapping area with the high-accuracy area | region of a conversion map. A2 is a second evaluation value indicating the overlapping area between the target circular area and the other circular areas extracted in step s3. A3 is a third evaluation value indicating the overlapping area between the target circular area and the low-accuracy area of the binarized map generated in step s2.

  For example, as shown in FIG. 27 described above, the threshold adjustment unit 7 is a binarized map in which all the circular regions extracted in step s3 are arranged (hereinafter, “binary map for evaluation value calculation”). Are used to obtain the first evaluation value A1, the second evaluation value A2, and the third evaluation value A3.

  Specifically, the threshold adjustment unit 7 obtains the number of a plurality of values constituting a portion overlapping the target circular region in the high accuracy region of the evaluation value calculation binarization map. Is a first evaluation value A1.

  Further, the threshold value adjusting unit 7 obtains the number of the plurality of values constituting the portion overlapping the other circular area in the target circular area in the evaluation value calculation binarization map, and calculates the obtained number. The second evaluation value A2. For example, assuming that four circular regions are extracted in step s3, the threshold value adjusting unit 7 determines a portion of the target circular region that overlaps with the other three circular regions in the evaluation value calculation binarization map. The number of the plurality of values constituting is obtained, and the obtained number is set as the second evaluation value A2.

  Then, the threshold adjustment unit 7 obtains the number of the plurality of values constituting the portion overlapping the target circular area in the low-accuracy area of the evaluation value calculation binarization map, and obtains the obtained number as the third number. Assume evaluation value A3.

  As can be understood from the above equation (2), the integrated evaluation value B for the target circular region is larger as the overlap area between the target circular region and the high-accuracy region of the binarized map is larger according to the first evaluation value A1. Become. The larger the circular area included in the high-accuracy area of the binarized map, the larger the overlap area between the circular area extracted from the binarized map and the high-accuracy area of the binarized map. Increases as the circular area included in the high-accuracy area of the binarized map increases.

  Further, the integrated evaluation value B for the target circular area becomes larger as the overlap area between the target circular area and the other circular areas extracted in step s3 becomes smaller by the second evaluation value A2. Therefore, the integrated evaluation value B increases when there is an independent circular area in the high accuracy area of the binarized map.

  Further, the integrated evaluation value B for the target circular area becomes larger as the overlap area between the target circular area and the low-accuracy area of the binarized map becomes smaller by the third evaluation value A3.

  Here, if the threshold value used in the generation of the binarization map is too small, each of the face images close to each other in the high accuracy region 51 of the binarization map 50 is shown in FIG. A plurality of corresponding regions 51a and 51b may be in contact with each other to form one independent region 511 that is nearly circular. That is, the high accuracy region 51 of the binarization map 50 may include one independent region 511 that is close to a circle corresponding to a plurality of face images. When a circular region is extracted from the high-accuracy region 51 of the binarization map 50 as described above, there is a possibility that one circular region 400 is extracted for the independent region 511 as shown in FIG.

  As described above, when the high-accuracy region of the binarization map includes an independent region 511 that is close to a circle and corresponds to a plurality of face images, it can be understood by comparing FIG. 28 with FIG. 27 described above. In addition, there is a possibility that the overlapping area between the circular region extracted from the high accuracy region of the binarized map and the low accuracy region of the binarized map may increase. Therefore, in this case, the third evaluation value A3 increases and the integrated evaluation value B decreases. In other words, if there are many independent circular regions corresponding to one face image in the high accuracy region of the binarized map, the circular region extracted from the high accuracy region of the binarized map and The overlapping area with the low accuracy region of the binarized map tends to decrease, and as a result, the integrated evaluation value B tends to increase as an independent circular region exists in the high accuracy region of the binarized map. It is in.

  As described above, the integrated evaluation value B increases as the circular area included in the high-accuracy area of the binarized map increases and increases as an independent circular area exists in the high-accuracy area of the binarized map. .

  When the threshold adjustment unit 7 obtains the integrated evaluation value B for each circular area extracted in step s3, the threshold adjustment unit 7 obtains the sum of the integrated evaluation values B and sets the obtained sum as the evaluation value for determination. Thereby, an evaluation value for determination indicating whether or not the threshold value temporarily set in step s1 is appropriate is obtained. The evaluation value for determination also increases as the circular area included in the high accuracy area of the binarized map increases, and increases as the independent circular area exists in the high accuracy area of the binarized map.

  When a plurality of determination evaluation values respectively corresponding to a plurality of stages of threshold values are obtained, the threshold adjustment unit 7 sets a threshold corresponding to the maximum value of the plurality of determination evaluation values in step s6. Decide on an appropriate threshold. When an appropriate threshold value is determined in this way, the binarization processing unit 5 binarizes the output value map using the appropriate threshold value, and generates a specifying binarization map. Thereby, the binarization map for specification is generated so as to include as many independent circular regions as possible while including as large a circular region as possible in the high accuracy region. The detection target image specifying unit 6 specifies a face image in the processing target image based on such a specifying binarization map, so that the detection target image specifying unit 6 responds with an output value map in the processing target image. A face image with a small detection accuracy value of a region to be identified can be specified, and each of a plurality of face images with close distances can be specified individually.

  When there are a plurality of maximum values in the plurality of determination evaluation values respectively corresponding to the threshold values in a plurality of stages, the threshold value corresponding to the maximum evaluation value among the plurality of determination evaluation values is the maximum value. It is preferable to determine the evaluation value for determination that is the minimum to an appropriate threshold value. This makes it easy to specify a face image having a small detection accuracy value of the corresponding region in the output value map in the processing target image.

  In the above example, when the integrated evaluation value B is calculated, the third evaluation value A3 indicating the overlapping area between the target circular region and the low-accuracy region of the binarized map is considered. The third evaluation value A3 may not be considered. That is, the integrated evaluation value B may be expressed by the following formula (3).

B = A1-A2 (3)
Even when the threshold adjustment unit 7 obtains the integrated evaluation value B using Equation (3), the high-accuracy region of the binarization map includes as large a circular region as possible, and an independent circular region. Can be determined so that the threshold value used in the generation of the binarization map is included.

  As described above, in the present embodiment, the threshold adjustment unit 7 has the first evaluation value A1 indicating the overlapping area between the circular region extracted from the high accuracy region of the binarized map and the high accuracy region. And the threshold value used in the generation of the binarization map based on the second evaluation value A2 indicating the overlapping area of the circular region and the other circular region extracted from the high-accuracy region Therefore, the threshold value used in the generation of the binarization map is adjusted so that the high-accuracy area of the binarization map includes as large a circular area as possible and includes an independent circular area as much as possible. can do. Therefore, the detection target image specifying unit 6 specifies the detection target image in the processing target image based on the binarized map generated using the adjusted threshold value, and thereby outputs the output value in the processing target image. A detection target image having a small detection accuracy value of a corresponding region on the map can be specified, and each of a plurality of detection target images having a short distance can be specified individually. Therefore, the detection accuracy for the detection target image is improved.

  In addition, as described above, the threshold value adjustment unit 7 performs not only the first evaluation value A1 and the second evaluation value A2, but also the third evaluation indicating the overlapping area of the circular region and the low accuracy region of the binarized map. By adjusting the threshold value based also on the value A3, it becomes possible to include more independent circular regions in the high accuracy region of the binarized map. Therefore, in the processing target image, each of a plurality of detection target images that are close to each other can be individually specified with high accuracy.

  Although the image detection apparatus 1 has been described in detail above, the above description is illustrative in all aspects, and the present invention is not limited thereto. For example, the detection target image may be an image other than a human face image. The various examples described above can be applied in combination as long as they do not contradict each other. And it is understood that the countless modification which is not illustrated can be assumed without deviating from the scope of the present invention.

DESCRIPTION OF SYMBOLS 1 Image detection apparatus 4 Map production | generation part 5 Binarization process part 6 Detection object image specific part 7 Threshold adjustment part 8 Circular area extraction part

Claims (6)

An image detection device for detecting a detection target image from a processing target image,
A map generation unit that generates a map indicating a distribution in the processing target image with respect to the accuracy value indicating the probability as the detection target image;
A binarization processing unit that binarizes the map using a threshold value to generate a binarized map;
A circular region from the first region in the binarized map generated using the threshold value corresponding to a region where the accuracy value is greater than or equal to the threshold value or larger than the threshold value in the map. An extraction unit for extracting
A first evaluation value indicating an overlap area between the first region and the circular region extracted from the first region of the binarized map by the extraction unit; and the first region from the first region by the circular region and the extraction unit. A threshold adjustment unit that adjusts the threshold based on a second evaluation value indicating an overlapping area with the other extracted circular regions;
A specifying unit that specifies the detection target image in the processing target image based on the first region of the binarized map generated by using the threshold adjusted by the threshold adjustment unit; An image detection apparatus comprising: The image detection apparatus according to claim 1,
The threshold adjuster is
A first evaluation value indicating an overlap area between the circular region extracted from the first region of the binarized map and the first region in the extraction unit;
A second evaluation value indicating an overlapping area between the circular region and another circular region extracted from the first region by the extraction unit;
A second region in the binarization map corresponding to the circular region and a region in the map where the accuracy value is less than or less than the threshold value used in the generation of the binarization map. An image detection apparatus that adjusts the threshold based on a third evaluation value indicating an overlapping area with a region. The image detection apparatus according to any one of claims 1 and 2,
The extraction unit is an image detection device that detects an edge of the binarized map and identifies a circular region in the first region of the binarized map by a Hough transform using the coordinates of the edge. The image detection apparatus according to any one of claims 1 to 3,
The image detection apparatus, wherein the detection target image is a human face image. A control program for controlling an image detection device that detects a detection target image from a processing target image,
In the image detection device,
(A) generating a map indicating a distribution in the processing target image with respect to an accuracy value indicating the probability as the detection target image;
(B) binarizing the map using a threshold value to generate a binarized map;
(C) From the partial area in the binarized map generated by using the threshold value corresponding to the area where the accuracy value is greater than or equal to the threshold value or larger than the threshold value in the map. Extracting a circular region;
(D) a first evaluation value indicating an overlapping area between the circular area extracted from the partial area of the binarized map and the partial area in the step (c); and the circular area and the step (c). Adjusting the threshold based on a second evaluation value indicating an overlapping area with another circular region extracted from the partial region;
(E) specifying the detection target image in the processing target image based on the partial region of the binarized map generated using the threshold value adjusted in the step (d); A control program to execute. An image detection method for detecting a detection target image from a processing target image,
(A) generating a map indicating a distribution in the processing target image with respect to an accuracy value indicating the probability as the detection target image;
(B) binarizing the map using a threshold value to generate a binarized map;
(C) From the partial area in the binarized map generated by using the threshold value corresponding to the area where the accuracy value is greater than or equal to the threshold value or larger than the threshold value in the map. Extracting a circular region;
(D) a first evaluation value indicating an overlapping area between the circular area extracted from the partial area of the binarized map and the partial area in the step (c); and the circular area and the step (c). Adjusting the threshold based on a second evaluation value indicating an overlapping area with another circular region extracted from the partial region;
(E) specifying the detection target image in the processing target image based on the partial region of the binarized map generated using the threshold value adjusted in the step (d); An image detection method comprising: