JP2005339389A - Picture processing method and picture processor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a picture processing method and a picture processor, with which existence of an object can surely be detected by a simple processing even if a gray scale picture including the object is image-picked up in a state where an illumination condition is nonuniform. <P>SOLUTION: A holding means 3 previously obtains the gray scale picture including the object of judgment and holds a template picture created by extracting a region including a feature of the object from the gray scale picture. A judging means 4 performs a correlation value operation of the gray scale picture for judgment, which is inputted from an A/D conversion means 2, and the template picture, and judges existence of the object in the gray scale picture for judgment from an operation result. When a shape of the object is almost symmetrical with a symmetrical element, the judging means 4 creates the gray scale picture which is almost symmetrical with the symmetrical element based on a reversed picture obtained by rotating the picture to a symmetrical position with the symmetrical element as a center and on the picture before reverse, and performs the correlation value operation by using the gray scale picture in a process for obtaining at least the template picture or the gray scale picture for judgment. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an image processing method and an image processing apparatus for determining the presence or absence of an object in an input grayscale image using template matching.

  Conventionally, an image processing method and an image processing apparatus for determining the presence or absence of an object in an input grayscale image using template matching have been used in various fields (see Patent Document 1). In general template matching, from the density values I (i, j) and T (i, j) for each pixel of the template image and the input grayscale image, the residual R and (formula 1) shown below (formula 1) The cross-correlation coefficient C shown in 2) (collectively referred to as “correlation value”) is obtained, and the degree of coincidence between both images is determined from the correlation value.

Incidentally, in recent years, methods and apparatuses for performing personal authentication using template matching as described above have been used in the security field. For example, as shown in FIG. 7A, a grayscale image (hereinafter simply referred to as “face image”) of a person who is permitted to enter a room is registered in advance as a template image TP, and a person who wishes to enter the room. When the face image V is input, the template image TP and the face image V are subjected to pattern matching to calculate a correlation value as shown in FIG. 7B, and the desired person enters and exits based on the correlation value. It is determined whether or not it matches with the person who is permitted.
JP-A-11-284997

  However, if template matching is performed using the grayscale image as it is, local spot light irradiation is performed when the grayscale image for creating the template image TP is captured and when the input grayscale image for determination is captured. And non-uniform changes in lighting conditions are added to the density value of the face image, making accurate pattern matching difficult.

  For example, as shown in FIG. 8A, when the illumination light is uniformly applied to the face of the person to be determined, the face image V of the person is placed on the center line (symmetric axis) L of the face passing through the nose. The image is axisymmetric with respect to the image, but as shown in FIG. 5B, when a local spot light is irradiated on the face of the person or the lighting condition becomes uneven, the face image V ′ at this time , The characteristics of the face to be determined are lost. Therefore, even if such a face image V ′ is compared with the template image TP, it is impossible to accurately determine whether or not the object exists.

  Here, the irradiation light is biased and the illumination condition is uneven is assumed to be a case where a strong light is exposed from one side of the face, and usually a side where the strong light is exposed is a bright image. On the other side, a portion where the irradiation light does not reach due to being blocked by the unevenness of the face or the like is generated, so that a dark image is obtained.

  Based on such a face image that is illuminated unevenly, a method for restoring an image that is illuminated uniformly is to detect the direction in which the illumination light is illuminated. Although there is a method of correcting the density value of the shadowed part due to the unevenness of the object (for example, Information Processing Society of Japan 2003-CVIM-139, P61 “Attitude and illumination by the appearance model construction on the 3D object surface Face matching that is robust to fluctuations)), the calculation process for correction is excessive, it is necessary to use a high-performance computer, and the calculation process takes a long time. There was a problem that it took.

  The present invention has been made in view of the above problems, and its object is to easily determine whether or not an object exists even when an input grayscale image including the object is captured under non-uniform illumination conditions. It is an object to provide an image processing method and an image processing apparatus that can be reliably detected by simple processing.

  In order to achieve the above object, the invention of claim 1 obtains a grayscale image including an object to be determined and extracts a template image created by extracting a region including the characteristics of the object from the grayscale image in advance. A grayscale image for determination for determining whether or not the object is included is obtained, and a correlation value calculation is performed between the grayscale image and the template image. In the process of obtaining at least one of the template image and the grayscale image for determination when the shape of the target object is substantially symmetric with respect to the symmetry element in determining whether or not the target object exists in Correction processing is performed so that at least one of the images becomes a gray image that is substantially symmetric with respect to the symmetry element, and a correlation value calculation is performed using the gray image after correction.

  According to a second aspect of the present invention, in the first aspect of the present invention, the correction processing obtains a density average value of two pixels at substantially symmetrical positions with respect to a symmetric element in at least one of the images, and this density average The process is characterized in that the value is a density value of two pixels.

  According to a third aspect of the present invention, in the first aspect of the invention, a reference value obtained by performing statistical calculation of the density value of each pixel in a grayscale image obtained by imaging an object under uniform illumination conditions is obtained in advance. In addition, the correction processing performs statistical calculation in two regions that are substantially symmetrical with respect to the symmetric element in at least one of the images, and the result of the statistical calculation is the density value of each pixel in the region closer to the reference value. The processing is to replace the density value of each pixel in the other region.

  According to a fourth aspect of the present invention, in the invention according to any one of the first to third aspects, the determination image is scanned with the template image to obtain a similarity region having the highest correlation value with the template image. An image having a size corresponding to the template image is cut out from the gray image for determination, and the cut image is corrected so as to become a gray image that is substantially symmetrical with respect to the symmetric element. Then, a correlation value calculation is performed, and the existence of an object is determined based on the calculation result.

  According to a fifth aspect of the present invention, in the invention according to any one of the first to fourth aspects, the template image is a grayscale image in which an image obtained by extracting a region including the characteristics of an object from a grayscale image is substantially symmetrical with respect to a symmetric element. It is an image obtained by correcting so that

  According to a sixth aspect of the present invention, in the invention according to any one of the first to fifth aspects, a grayscale image for determination is scanned with a template image to obtain a first similarity region having the highest correlation value with the template image. A process of performing correlation value calculation in the first similarity region, and an image having a size corresponding to the template image is cut out from the grayscale image for determination in the first similarity region, and the cut-out image is substantially symmetrical with respect to the symmetric element. The correction is performed so that the image becomes a grayscale image, the correlation value is calculated between the corrected grayscale image and the template image, and the template image is corrected so as to become a grayscale image that is substantially symmetrical with respect to the symmetric element. The gray image is used as a new template image, and the inside of the gray image for determination is scanned with the changed template image to obtain a second similar region having the highest correlation value with the changed template image. Processing for performing correlation value calculation in the region, and cutting out an image having a size corresponding to the template image after the change from the grayscale image for determination in the second similar region, and the cut-out image is substantially symmetrical with respect to the symmetric element Correction is performed so as to obtain a grayscale image, and a correlation value calculation is performed on the corrected grayscale image and the modified template image, and the result of at least one correlation value calculation exceeds a predetermined threshold value. And determining that the object is present.

  The invention according to claim 7 is a holding means for obtaining a grayscale image including a target object to be determined in advance, extracting a region including the characteristics of the target object from the grayscale image, and holding the template image; An image input means for inputting a gray image for determination for determining the presence or absence of the image, and a correlation value calculation between the gray image for determination and the template image are performed, and from the calculation result, the object in the gray image for determination is calculated. Determining means for determining presence / absence, and when the shape of the object is substantially symmetrical with respect to the symmetrical element, the determining means selects at least one of the template image and the gray image for determination. In the obtaining process, correction processing is performed so that the image becomes a gray image that is substantially symmetric with respect to the symmetric element, and correlation value calculation is performed using the gray image after correction.

  According to the first aspect of the present invention, the template image including the object or the grayscale image for determination is provided because the illumination condition is not uniform even though the shape of the object is substantially symmetrical with respect to the symmetry element. Becomes an asymmetric image with respect to the symmetric element, and even if the feature of the object is lost, at least one of the template image and the grayscale image for determination is substantially symmetric with respect to the symmetric element. Since correction processing is performed to obtain a grayscale image, and the correlation value calculation is performed using the corrected grayscale image, the feature of the target object included in the grayscale image is restored to ensure the presence or absence of the target object. An image processing method that can be determined can be provided.

  According to the second aspect of the present invention, in at least one of the images, the average density value of two pixels located substantially symmetrically with respect to the symmetrical element is obtained, and the average density value is obtained as the density value of the two pixels. By doing so, since a gray image that is substantially symmetric with respect to the symmetric element is created, it is possible to provide an image processing method capable of creating a substantially symmetric image with respect to the symmetric element with a simple process.

  According to the invention of claim 3, a reference value obtained by performing statistical calculation of the density value of each pixel in an arbitrary comparison region in a grayscale image obtained by imaging an object under uniform illumination conditions is obtained in advance. In at least one of the images, statistical calculation is performed in two regions that are substantially symmetric with respect to the symmetric element, and the result of the statistical calculation is the density value of each pixel in the region closer to the reference value. By replacing the density value of each pixel in the region, a gray image that is substantially symmetric with respect to the symmetric element is created. Therefore, an image processing method that can create a substantially symmetric image with respect to the symmetric element with simple processing. Can be provided.

  According to the fourth aspect of the present invention, the density image for determination is scanned with the template image to obtain a similar region having the highest correlation value, and the image in the similar region is converted into a gray image that is substantially symmetric with respect to the symmetric element. Since the correlation value is calculated between the corrected gray image and the template image and the presence or absence of the target is determined based on the calculation result, the illumination when the gray image for determination is captured It is possible to provide an image processing method capable of reliably determining the presence or absence of an object even if the conditions are not uniform.

  According to the invention of claim 5, the template image is an image obtained by correcting an image obtained by extracting a region including the characteristics of the object from the grayscale image so as to become a grayscale image substantially symmetric with respect to the symmetric element. Even if the characteristics of the object are impaired because the gray image for template creation is captured under non-uniform illumination conditions, the gray image for the template is corrected so that it becomes a substantially symmetric gray image with respect to the symmetric element. Thus, it is possible to provide an image processing method that can restore the characteristics of an object and can reliably determine the presence or absence of the object.

  According to the invention of claim 6, the processing for performing the correlation value calculation between the gray image of the first similar region and the template image, and the image of the first similar region so as to become a gray image substantially symmetrical with respect to the symmetric element. Processing for calculating a correlation value between the corrected image and the template image, and calculating a correlation value between the corrected template image corrected to become a substantially symmetric image with respect to the symmetric element and the grayscale image of the second similar region And a process of calculating a correlation value between the corrected template image and the image of the second similar region corrected so as to become a gray image that is substantially symmetric with respect to the symmetric element. Among the calculation results of processing, if at least one calculation result exceeds a predetermined threshold value, it is determined that an object is present, so that the illumination conditions when capturing a grayscale image or template image for determination are uniform. In Also be nonuniform, it is possible to provide a reliably determinable image processing method the presence or absence of the object.

  According to the invention of claim 7, the template image including the object or the gradation image for determination is included because the illumination condition is not uniform although the shape of the object is substantially symmetrical with respect to the symmetry element. Even if the image becomes asymmetric with respect to the symmetric element and the characteristics of the target object are lost, the determination means selects at least one of the template image and the grayscale image for determination with respect to the symmetric element. Therefore, the correlation value calculation is performed using the corrected grayscale image, and the feature of the target object included in the grayscale image is restored. It is possible to provide an image processing apparatus capable of reliably determining the presence or absence.

(Embodiment 1)
FIG. 1 shows an embodiment in which an image processing apparatus for carrying out the image processing method of the present invention is applied to an entrance management apparatus that permits entry only to an authenticated person. ) Including an image pickup means 1 such as a CCD camera, an A / D conversion means 2 for converting an analog image signal picked up by the image pickup means 1 into a digital signal, and an object to be determined in advance. A holding unit 3 for holding a template image created by obtaining a grayscale image including (for example, a person permitted to enter the room) and extracting a region including the feature of the object from the grayscale image, and an A / D conversion unit 2 is used to calculate a correlation value between the grayscale image for determination input from 2 and the template image, and based on the calculation result, the presence or absence of an object in the grayscale image for determination, in other words, the face of the imaged person H is a template. And a determining means 4 or not it matches the face of pre-registered person as an image. Note that at least the holding unit 3 and the determination unit 4 are configured by mounting software (program) for causing a general-purpose computer to execute processing in each unit, for example.

  When an image including an object is picked up by the image pickup means 1 and an analog image signal picked up by the image pickup means 1 is converted into a digital signal by the A / D conversion means 2 and inputted to the determination means 4, the determination means 4 calculates a correlation value between the input grayscale image for determination input from the A / D conversion means 2 and the template image held in the holding means 3, and determines the degree of coincidence of the face of the person to be determined. .

  First, the determination unit 4 places a template image at the initial position of the input grayscale image for determination (for example, the position of the upper left corner of the image), and each pixel of the template image and the input grayscale image (hereinafter referred to as “partial image”). The difference between the pixel values (density values) is obtained from each pixel of the pixel and the absolute value of the difference is set as the correlation value a. Then, when the determination unit 5 obtains the correlation value a for all the pixels in the partial image, a new value obtained by dividing the sum of the correlation values a by the number of all pixels included in the partial image is newly set as the correlation value b. Define.

  The correlation value b obtained in this way decreases as the degree of coincidence increases, and increases as the degree of coincidence decreases. Therefore, if the correlation value b is equal to or smaller than the predetermined threshold value, the determination unit 4 determines that the partial image matches the template image, and determines that the target person H matches the person registered in advance. The result is output and the process is terminated. If the partial image does not match the template image because the correlation value b is larger than the predetermined threshold value, the determination unit 4 determines whether the edge of the partial image matches the horizontal edge of the input image. In other words, it is determined whether or not the above determination processing is completed for all the pixels in the horizontal direction at the same position in the vertical direction. If they do not match, the template image is set in the horizontal direction with respect to the input image. The determination process is performed with respect to a new partial image with a shift of one pixel. When the edge of the partial image matches the edge of the input image in the horizontal direction, the determination unit 4 determines whether the edge of the partial image matches the edge of the input image in the vertical direction, that is, in the horizontal direction. It is determined whether or not the above determination processing has been completed for all the pixels in the vertical direction at the same position in FIG. 3. If they do not match, the template image is shifted by one pixel in the vertical direction with respect to the input image to newly The above determination processing is performed with respect to a partial image. If the edge of the partial image coincides with the edge of the input image in the vertical direction, the determination unit 4 does not have an image that matches the template image in the input image, that is, the target person H is registered in advance. A determination result indicating that the person does not match is output, and the process ends.

  By the way, the face of the person to be determined has left-right symmetry and is line symmetric with respect to the center line (symmetry axis) of the face passing through the nose. As shown in FIG. If local spot light irradiation or non-uniform changes in illumination conditions are added to the density value in TP, the symmetry of the face image will be lost, the characteristics of the target will be lost, and pattern matching can be performed accurately. There is a possibility of disappearing. Similarly, if a local spotlight or nonuniform change in illumination conditions is added to the density value of the face image when capturing an input grayscale image for determination, the symmetry of the face image is lost and the target image is lost. Since the feature of the object is lost, there is a possibility that pattern matching cannot be performed accurately.

  Therefore, in the present embodiment, in the process of obtaining the template image and the gradation image for determination, the determination unit 4 creates inverted images obtained by rotating the template image and the gradation image for determination to symmetrical positions around the symmetric elements. Then, based on the inverted image and the original image, correction processing is performed so that the original image becomes a gray image that is substantially symmetric with respect to the symmetry element, and the presence / absence of the object is determined using the corrected gray image. Is determined.

  FIG. 2A is an example of a template image TP. Since the template image TP is irradiated with light from the left side of the person's face at the time of imaging, the template image TP is a region on the right side compared to the region on the left side across the symmetry axis L. Is darker. At this time, the determination means 4 obtains a gray image symmetric with respect to the symmetry axis L by rotating the template image TP around the symmetry axis L by 180 degrees, and uses this gray image and the gray image V1 before inversion. A grayscale image that is substantially symmetric with respect to the symmetry axis L is created. For example, the average value of the density value of each pixel of the inverted image and the density value of the corresponding pixel of the template image TP before the inversion is actually calculated. The density value of the corresponding pixel of the template image TP ′ used for the correlation value calculation is used. That is, assuming that the pixel values of two pixels symmetrical with respect to the symmetry axis in the template image TP are a1 and b1, the pixel values A1 and B2 of the corresponding two pixels of the template image TP ′ used for the actual correlation value calculation are It is shown by the following (Formula 3).

A1 = A2 = (a1 + b1) / 2 (Formula 3)
As described above, in the determination unit 4, the average value of two pixels located symmetrically with respect to the symmetry axis L in the template image TP is used as the average value of the two pixels in the template image TP ′ used for the actual correlation value calculation. Since the pixel value (density value) is used, when the template image TP held in advance is irradiated with spot light only on one side of the face, the density value of each pixel becomes asymmetric with respect to the symmetry axis. However, it is possible to create a template image TP ′ with no bias in brightness by substantially averaging the pixel values of the bright and dark regions in the template image TP, and use this template image TP ′ for actual determination. Thus, the presence / absence of the object can be accurately determined.

  In the above description, the case where the template image TP is corrected to the template image TP ′ substantially symmetric with respect to the symmetry axis has been described as an example. However, the same processing as described above is performed when obtaining a grayscale image for determination. By doing so, the grayscale image for determination can be changed to a grayscale image that is substantially symmetrical with respect to the symmetry axis. Note that it is not necessary to obtain an average value of density values for all pixels of the grayscale image with the corresponding pixels in the inverted image, and this average value need not be used as the pixel value of the grayscale image after correction. If the correlation value calculation is performed only in the characteristic area (for example, the area of eyes, eyebrows, mouth, nose, etc. in the case of a human face) to identify the target object, the symmetry axis only in this characteristic area The correction processing similar to that described above may be performed so as to obtain a substantially target image, and the correlation value calculation may be performed on the corrected grayscale image.

(Embodiment 2)
A second embodiment of the present invention will be described with reference to FIG. Since the basic configuration of the image processing apparatus is the same as that of the first embodiment, common components are denoted by the same reference numerals, and description thereof is omitted.

  In Embodiment 1 described above, when the shape of the object is a line-symmetric shape with respect to the symmetry axis, the template image and the grayscale image for determination are obtained by inverting them to positions that are symmetric with respect to the symmetry axis. The average value of the density value of each pixel of the inverted image and the density value of the corresponding pixel of the image before inversion is used as the density value of each pixel of the grayscale image used for actual correlation value calculation. In this case, the density value of each pixel of the grayscale image used for the correlation value calculation is the density value of the corresponding pixel of the inverted image or the density value of the corresponding pixel of the grayscale image before inversion.

  That is, the determination unit 4 obtains a face image obtained by imaging under uniform illumination conditions with no bias in brightness, and each of the two left and right comparison regions divided with the symmetry axis of the face image as a boundary. The statistical calculation of the pixel values of all the pixels included in each comparison area is performed, and the calculation results (for example, average value and variance) are stored in advance as reference values.

  Then, in the determination unit 4, in the process of obtaining the template image and the grayscale image for determination, the template image and the grayscale image for determination are inverted to a symmetrical position with respect to the symmetric element, and the reverse image obtained before the reverse image The density value of each pixel is statistically calculated in the corresponding comparison area, and the density value of each pixel in the comparison area of the image whose calculation result is closer to the reference value is used for the correlation value calculation. The density value of each pixel in FIG. For example, the determination unit 4 includes two regions that are substantially symmetric with respect to the symmetry axis (symmetric element) of each image, in this case, each of the two regions (comparison regions) divided right and left with the symmetry axis as a boundary. Statistical calculation of the pixel values of all pixels included in (average value calculation) is performed, and the area where the result of this statistical calculation (average value) is closer to the reference value obtained in advance is adopted. A density image used for actual correlation value calculation is obtained by performing correction processing that replaces the pixel value of the corresponding pixel in the other region with the pixel value of the pixel.

  FIG. 3A is an example of the template image TP. Since the template image TP is irradiated with light from the left side of the person's face at the time of imaging, the template image TP is on the right side of the left side region W1 with the symmetry axis L in between. The area W2 is darker. At this time, the determination unit 4 divides the template image TP into two left and right regions W1 and W2 with the symmetry axis L as a boundary, and obtains an average value of the pixel values of all the pixels in each of the two left and right regions W1 and W2. The area closer to the reference value obtained in advance is adopted, and the pixel value of each pixel in the opposite area is replaced with the pixel value of all the pixels in the adopted area. For example, if the average pixel value of the left region W1 with respect to the symmetry axis L is closer to the reference value, the pixel value of each pixel in the right region W2 is replaced with the pixel value of the corresponding pixel in the left region W1. Thus, a grayscale image used for correlation value calculation is obtained. That is, the pixel values of two pixels that are symmetric with respect to the symmetry axis L in the template image TP ′ after the change are A2 and B2, and the pixel values of the two pixels in the original template image TP corresponding to these two pixels. In this case, A2 = B2 = a2. If the average pixel value of the right region W2 with respect to the symmetry axis L is closer to the reference value, the pixel value of each pixel in the left region W1 is replaced with the pixel value of the corresponding pixel in the right region W1. Thus, a grayscale image used for correlation value calculation is obtained. In this case, A2 = B2 = b2.

  As described above, the determination unit 4 obtains an average value of pixel values in an arbitrary comparison region in a grayscale image captured under ideal illumination conditions, grasps this average value in advance as a reference value, and then When a template image is input, the average value of the pixel values in the corresponding comparison area is obtained for each of the template image and the inverted image obtained by inverting the template image with respect to the symmetrical element, and the one closer to the previously obtained reference value Since the pixel value of each pixel in the comparison area of the image is the pixel value of each pixel in the comparison area of the image used for the actual correlation value calculation, spot light is irradiated only on one side of the face image, etc. Even when the density value of each pixel is asymmetric with respect to the symmetry axis, the pixel values of the bright area and the dark area in the template image TP are approximately averaged, so that the template image has no bias in brightness. 'Can be created, the template image TP' P by using the actual correlation value calculation, it is possible to determine presence or absence of the object accurately.

  In the above description, the case where the template image TP is corrected to the template image TP ′ that is substantially symmetric with respect to the symmetry axis has been described as an example. However, the same correction processing as described above is performed when obtaining a grayscale image for determination. By performing the above, it is possible to change the grayscale image for determination into a grayscale image that is substantially symmetric with respect to the symmetry axis.

  In the present embodiment, the grayscale image is divided into two left and right regions sandwiching the symmetry axis L, the average value of each pixel in the region is obtained for each region, and the region closer to the previously obtained reference value is adopted. The pixels in the opposite area are replaced with the pixels in the selected area, but the characteristic areas for identifying the object in the grayscale image (for example, eyes, eyebrows, mouths, If the correlation value calculation is performed only in the region such as the nose, the above processing may be executed only in the feature region. That is, in a grayscale image captured in an ideal illumination environment, an average value of density values of each pixel in a region corresponding to the eye portion is obtained in advance, and a template image or an input grayscale image for determination is obtained. In the process of obtaining, the average value of the density value of each pixel in the two areas corresponding to the eye portion in these images and symmetrical with respect to the symmetry axis is obtained, and an area close to the previously obtained average value is obtained. Alternatively, the density value of each pixel in this region may be replaced with the density value of each pixel in this region.

  Further, in each of the above-described embodiments, the determination unit 4 applies the symmetric element to the symmetric element from the inverted image obtained by inverting both the template image TP and the grayscale image for determination to a symmetric position with respect to the symmetric element and the image before the inversion. Image, and the correlation value calculation is performed using this image. However, any one of the template image TP or the grayscale image for determination taken under non-uniform illumination conditions An image that is substantially symmetric with respect to the symmetric element is created from the inverted image and the image before being inverted, and the correlation value calculation may be performed using this image.

(Embodiment 3)
A third embodiment of the present invention will be described with reference to FIG. Since the basic configuration of the image processing apparatus is the same as that of the first embodiment, common components are denoted by the same reference numerals, and description thereof is omitted.

  In conventional template matching, the input grayscale image is scanned with a template image prepared in advance, the correlation value is calculated at each position, and it is assumed that there is an object at the position where the correlation value is maximized. The presence / absence of the object was determined using the value.

  Here, when the input grayscale image for determination or the template image is captured under non-uniform illumination conditions, template matching is performed and the presence or absence of an object is determined using the maximum correlation value. In a grayscale image captured under illumination conditions, the symmetry with respect to the symmetry axis is lost, so the correlation value tends to be low, and there is an error that the object does not exist even though the object exists. There was a possibility of making a judgment.

  Therefore, in the present embodiment, the determination unit 4 scans the input grayscale image for determination using the template image held in the holding unit 3 to search for the first similar region having the highest correlation value, and the correlation is performed in this similar region. After performing the value calculation, the template image is converted into a gray image that is substantially symmetric with respect to the symmetrical element by the method described in the first embodiment, and the input gray image for determination is scanned again using the converted template image. The second similar region having the highest correlation value is searched, the correlation value is calculated again in the second similar region, and the presence / absence of the object is determined based on the calculation result of the two correlation value calculations. .

  Here, the determination process by the determination means 4 is demonstrated based on the flowchart of FIG. First, the determination unit 4 places a template image at an initial position (for example, the position of the upper left corner) of the input image (S1), and between each pixel of the template image and each pixel of the input image (partial image) that overlaps the template image. To obtain the difference between the pixel values (density values) and set the absolute value of the difference as the correlation value a for each pixel. Then, when the correlation value a is obtained for all the pixels in the partial image, a sum of these correlation values a divided by the total number of pixels included in the partial image is newly defined as a correlation value b. The correlation value b at the position is stored (S2). When the calculation of the correlation value b is completed, the determination unit 4 determines whether or not the edge of the partial image coincides with the edge of the input image in the horizontal direction, that is, for all the pixels in the horizontal direction at the same position in the vertical direction. It is determined whether or not the determination process has been completed (S3). If they do not match, the template image is shifted by one pixel in the horizontal direction with respect to the input image, and the determination is performed with a new partial image. Processing is performed (S4). When the edge of the partial image matches the edge of the input image in the horizontal direction, the determination unit 4 determines whether the edge of the partial image matches the edge of the input image in the vertical direction, that is, in the horizontal direction. In step S5, it is determined whether or not the above-described determination processing has been completed for all the pixels in the vertical direction at the same position in FIG. The determination process is performed with respect to a new partial image by shifting (S6). If the edge of the partial image matches the edge of the input image in the vertical direction, the determination unit 4 determines that the determination process has been completed between the template image and the partial image at all positions of the input image. Among the correlation values b at all positions, the largest correlation value A and the position of the template image at that time are stored (S7).

  Thereafter, the determination unit 4 converts the template image into a gray image that is substantially symmetrical with respect to the symmetric element using the method described in the first embodiment, and uses this gray image as a new template image (S8). The correlation values between the new template image and the partial image are obtained at all positions of the input image by performing the same processing as S1 to S6 (S9 to S14), and among the correlation values at all positions, The largest correlation value B and the position of the template image at that time are stored (S15).

  When the template matching using the pre-conversion and post-conversion template images is completed, the determination unit 4 compares the two correlation values A and B with a predetermined threshold value, and either one of the correlation values When A and B exceed the threshold value, it is determined that there is an object, and the determination result and template information indicating whether the matched template image is before conversion or after conversion are output (S16). Note that the determination unit 4 may determine that the object exists only when both of the two correlation values A and B exceed the threshold value, and can reliably determine whether the object exists.

  In the present embodiment, the template image held in the holding unit 3 is converted into an image that is substantially symmetric with respect to the symmetry axis by the method described in the first embodiment to obtain a new template image. The image may be converted into a substantially symmetric image with respect to the symmetry axis by the method described above.

(Embodiment 4)
A fourth embodiment of the present invention will be described with reference to FIG. Since the basic configuration of the image processing apparatus is the same as that of the first embodiment, common components are denoted by the same reference numerals, and description thereof is omitted.

  As described in the third embodiment, when an input grayscale image or template image for determination is captured under non-uniform illumination conditions, template matching is performed and the presence or absence of an object is determined using the maximum correlation value. In contrast, in a grayscale image captured under non-uniform illumination conditions, since the symmetry with respect to the symmetry axis is impaired, the correlation value tends to be low, and the object is present despite the presence of the object. There was a possibility of making a false judgment that it did not exist.

  Therefore, in the present embodiment, the determination unit 4 scans the input grayscale image for determination with the template image to search for a similar region having the highest correlation value, performs a correlation value calculation in this similar region, and then performs the similar region. Then, an image having a size corresponding to the template image is cut out, and the cut image is converted into a gray image that is substantially symmetric with respect to the symmetric element by the method described in the first embodiment, and the gray image and the template image are correlated again. Value calculation is performed, and the presence or absence of an object is determined based on the calculation result of two correlation value calculations.

  Here, the determination process by the determination means 4 is demonstrated based on the flowchart of FIG. First, the determination unit 4 places a template image at an initial position (for example, the position of the upper left corner) of the input image (S21), and between each pixel of the template image and each pixel of the input image (partial image) overlapping therewith. To obtain the difference between the pixel values (density values) and set the absolute value of the difference as the correlation value A for each pixel. Then, when the correlation value a is obtained for all the pixels in the partial image, a sum of these correlation values a divided by the total number of pixels included in the partial image is newly defined as a correlation value b. The correlation value b at the position is stored (S22). When the calculation of the correlation value b is completed, the determination unit 4 determines whether or not the edge of the partial image coincides with the edge of the input image in the horizontal direction, that is, for all the pixels in the horizontal direction at the same position in the vertical direction. It is determined whether or not the determination process has been completed (S23). If they do not match, the determination is performed with respect to a new partial image by shifting the template image by one pixel in the horizontal direction with respect to the input image. Processing is performed (S24). When the edge of the partial image matches the edge of the input image in the horizontal direction, the determination unit 4 determines whether the edge of the partial image matches the edge of the input image in the vertical direction, that is, in the horizontal direction. In step S25, it is determined whether or not the above-described determination processing has been completed for all the pixels in the vertical direction at the same position in FIG. The determination process is performed with respect to a new partial image by shifting (S26). If the edge of the partial image matches the edge of the input image in the vertical direction, the determination unit 4 determines that the determination process has been completed between the template image and the partial image at all positions of the input image. Of the correlation values at all positions, the largest correlation value C and the position of the template image at that time are stored (S27).

  Next, the determination unit 4 cuts out an image having a size corresponding to the template image from the input image in the similar region having the maximum correlation value, and the cut image is substantially symmetric with respect to the symmetric element by the method described in the first embodiment. After the conversion to a clear gray image (S28), the correlation value calculation is performed again between the converted gray image and the template image to obtain the correlation value D (S29).

  When the above processing is completed, the determination unit 4 compares the two correlation values C and D with a predetermined threshold value, and if any one of the correlation values C and D exceeds the threshold value, It is determined that there is an object, and input image information information indicating the determination result and whether the matched input image is the one before conversion or after conversion is output (S30). Note that the determination means 4 may determine that the object exists only when both of the two correlation values C and D exceed the threshold value, and can reliably determine whether the object exists.

  In this embodiment, the presence / absence of the object is determined from the result of the correlation value calculation twice. However, the correlation value calculation is performed only once, and the determination unit 4 scans the input grayscale image for determination with the template image. Then, a similar region having the highest correlation value is searched, an image having a size corresponding to the template image is cut out in the similar region, and the cut-out image is substantially symmetric with respect to a symmetric element by the method described in the first embodiment. It may be converted into a grayscale image, a correlation value calculation is performed between the grayscale image and the template image, and the presence or absence of an object may be determined based on the calculation result.

  In the present embodiment, the grayscale image for determination is converted into an image that is substantially symmetric with respect to the symmetry axis by the method described in the first embodiment, and the correlation value is calculated between the converted image and the template image. However, the image may be converted into a substantially symmetric image with respect to the symmetry axis by the method described in the second embodiment.

(Embodiment 5)
A fifth embodiment of the present invention will be described with reference to FIG. Since the basic configuration of the image processing apparatus is the same as that of the first embodiment, common components are denoted by the same reference numerals, and description thereof is omitted.

  As described in the third embodiment, template matching is performed when an input grayscale image or template image for determination is captured under non-uniform illumination conditions, and the presence or absence of an object is determined using the maximum correlation value. Then, in a grayscale image captured under non-uniform illumination conditions, since the symmetry with respect to the symmetry axis is lost, the correlation value tends to be low, and the target object exists despite the presence of the target object. There was a possibility of making a false judgment that there was no.

  Therefore, in the present embodiment, the determination unit 4 scans the input grayscale image for determination with the template image, searches for the first similar region having the highest correlation value, calculates the correlation value in the first similar region, An image having a size corresponding to the template image is cut out in the first similarity region, a grayscale image obtained by converting the cutout image by the method described in the first embodiment is obtained, and the correlation between the converted grayscale image and the template image is obtained. After performing the value calculation, the template image is converted by the method described in the first embodiment, and the input gradation image for determination is scanned using the converted template image to search for the second similar region having the highest correlation value. Then, the correlation value is calculated in the second similar region, and an image having a size corresponding to the template image after the change is cut out in the second similar region, and the cut out image is described in the first embodiment. The grayscale image converted by the above method is obtained, and the correlation value calculation between the converted grayscale image and the modified template image is performed. Based on the calculation result of the correlation value calculation four times in total, the presence / absence of the object is determined. Deciding.

  Here, the determination process by the determination means 4 is demonstrated based on the flowchart of FIG. First, the determination unit 4 places a template image at an initial position (for example, the position of the upper left corner) of the input image (S31), and between each pixel of the template image and each pixel of the input image (partial image) overlapping therewith. To obtain the difference between the pixel values (density values) and set the absolute value of the difference as the correlation value a for each pixel. Then, when the correlation value a is obtained for all the pixels in the partial image, a sum of these correlation values a divided by the total number of pixels included in the partial image is newly defined as a correlation value b. The correlation value b at the position is stored (S32). When the calculation of the correlation value b is completed, the determination unit 4 determines whether or not the edge of the partial image coincides with the edge of the input image in the horizontal direction, that is, for all the pixels in the horizontal direction at the same position in the vertical direction. It is determined whether or not the determination process has been completed (S33). If they do not match, the template image is shifted by one pixel in the horizontal direction with respect to the input image, and the determination is performed with a new partial image. Processing is performed (S34). When the edge of the partial image matches the edge of the input image in the horizontal direction, the determination unit 4 determines whether the edge of the partial image matches the edge of the input image in the vertical direction, that is, in the horizontal direction. In step S35, it is determined whether or not the above-described determination processing has been completed for all pixels in the vertical direction at the same position in FIG. The determination process is performed with respect to a new partial image by shifting (S36). If the edge of the partial image matches the edge of the input image in the vertical direction, the determination unit 4 determines that the determination process has been completed between the template image and the partial image at all positions of the input image. Of the correlation values at all positions, the largest correlation value E and the position of the template image at that time are stored (S37).

  Next, the determination unit 4 cuts out an image having a size corresponding to the template image from the input image in the first similar region having the maximum correlation value, and converts the grayscale image obtained by converting the cut-out image by the method described in the first embodiment. Obtaining (S38), the correlation value is calculated between the converted gray image and the template image to obtain the correlation value F (S39).

  When the above processing is completed, the determination unit 4 converts the template image into a substantially symmetric image with respect to the symmetric element using the method described in the first embodiment, and uses this image as a new template image (S40). The same processing as in S31 to S36 described above is performed to obtain correlation values between the changed template image and partial images at all positions of the input image (S41 to S46), and correlation values at all positions Among them, the largest correlation value G and the position (second similarity region) of the new template image at that time are stored (S47).

  Thereafter, the determination unit 4 cuts out an image having a size corresponding to the template image from the input image in the second similar region where the correlation value is maximized, and converts the grayscale image obtained by converting the cut-out image by the method described in the first embodiment. Obtaining (S48), the correlation value is calculated between the converted gray image and the template image to obtain the correlation value H (S49).

  When the four correlation value operations are completed, the determination unit 4 compares the four correlation values E to H with a predetermined threshold value, and any one correlation value exceeds the threshold value. And the determination result and input image information and template information indicating whether the matched input image and template image are before conversion or after conversion are output (S50). Note that the determination unit 4 may determine that the object exists only when all of the four correlation values E to H exceed the threshold value, and can reliably determine whether the object exists.

  In the present embodiment, the grayscale image and template image for determination are converted into an image that is substantially symmetric with respect to the symmetry axis by the method described in the first embodiment. Although the correlation value calculation is performed between the template image after the change and the template image after the change, the grayscale image for determination and the template image may be converted into an image substantially symmetric with respect to the symmetry axis by the method described in the second embodiment. good.

  In each of the above-described embodiments, the case where the object to be determined is a human face having a line-symmetric shape with respect to the symmetry axis has been described. However, the shape of the object is linear with respect to the symmetry axis. It is not intended to be limited to a symmetric shape, and when the object is in a point-symmetric shape with respect to the center of symmetry (for example, a diamond-shaped object), if the illumination conditions are not uniform A reversed image is created by rotating a part of the image of the object to a position symmetric with respect to the center of symmetry, and a grayscale image that is substantially point-symmetric with respect to the center of symmetry from this reversed image and the image before the reversal. And a correlation value calculation may be performed using this grayscale image.

1 is a block diagram illustrating an image processing apparatus according to a first embodiment. (A) (b) is explanatory drawing of the grayscale image used for the same as the above. (A) (b) is explanatory drawing of the grayscale image used for Embodiment 2. FIG. 10 is a flowchart for explaining the operation of the image processing apparatus according to the third embodiment. 10 is a flowchart for explaining the operation of the image processing apparatus according to the fourth embodiment. 10 is a flowchart for explaining the operation of the image processing apparatus according to the fifth embodiment. (A) (b) is explanatory drawing explaining the conventional pattern matching. (A) is an example figure of the grayscale image imaged on uniform illumination conditions, (b) is an example figure of the grayscale image imaged on nonuniform illumination conditions.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Imaging means 2 A / D conversion means 3 Holding means 4 Judgment means

Claims (7)

  Obtaining a grayscale image including the target object to be determined, extracting a region including the characteristics of the target object from the grayscale image, holding in advance a template image, and whether or not the target object is included Obtaining a grayscale image for determination for determination, performing a correlation value calculation between the grayscale image and the template image, and determining the presence or absence of the object in the grayscale image for determination from this calculation result, In the process of obtaining at least one of the template image and the gray image for determination when the shape of the object is substantially symmetrical with respect to a symmetric element, the at least one image is obtained. An image processing method comprising: performing correction processing so that a gray image is substantially symmetric with respect to the symmetric element, and performing correlation value calculation using the corrected gray image.   In the correction process, in at least one of the images, a density average value of two pixels located substantially symmetrical with respect to the symmetry element is obtained, and this density average value is set as the density value of the two pixels. The image processing method according to claim 1, wherein the image processing method is processing.   A reference value obtained by performing a statistical calculation of the density value of each pixel in a grayscale image obtained by imaging the object under uniform illumination conditions is obtained in advance, and the correction process is performed by the at least one of the above correction processes. In the image, statistical calculation is performed in two areas that are substantially symmetric with respect to the symmetric element, and the result of the statistical calculation is the density value of each pixel in the area closer to the reference value, and the density of each pixel in the other area The image processing method according to claim 1, wherein the process is a process of replacing a value.   The determination image is scanned with the template image to obtain a similar region having the highest correlation value with the template image, and the similarity region has a size corresponding to the template image from the determination gray image. The image is cut out, the cut out image is corrected so as to be a gray image that is substantially symmetric with respect to the symmetric element, and a correlation value calculation is performed between the corrected gray image and the template image to obtain a calculation result. The image processing method according to claim 1, wherein the presence or absence of the object is determined.   The template image is an image obtained by correcting an image obtained by extracting a region including the feature of the object from the grayscale image so as to become a grayscale image substantially symmetric with respect to the symmetric element. The image processing method according to any one of claims 1 to 4.   Scanning the gray image for determination with the template image, obtaining a first similarity region having the highest correlation value with the template image, and performing a correlation value calculation in the first similarity region; and the first similarity An image having a size corresponding to the template image is cut out from the gray image for determination in the region, and the cut image is corrected so as to become a gray image substantially symmetric with respect to the symmetric element. Processing for calculating a correlation value between the image and the template image, correcting the template image so as to become a gray image that is substantially symmetrical with respect to a symmetric element, setting the corrected gray image as a new template image, and performing the determination The gray image for scanning is scanned with the template image after the change, and the second similar region having the highest correlation value with the template image after the change is obtained, and the correlation is performed in the second similar region. A process of performing an operation, and an image having a size corresponding to the template image after change is cut out from the grayscale image for determination in the second similar region, and the cut-out image is substantially symmetrical with respect to the symmetric element. A correction is performed so that a grayscale image is obtained, and a correlation value calculation is performed on the corrected grayscale image and the modified template image, and at least one of the correlation value calculation results in a predetermined threshold value. The image processing method according to any one of claims 1 to 5, wherein if it exceeds, it is determined that the object exists. Obtaining a grayscale image including a target object to be determined in advance, extracting a region including the characteristics of the target object from the grayscale image, holding means for holding a template image, and determining whether the target object exists Image input means for inputting a grayscale image for determination, and a correlation value calculation between the grayscale image for determination and the template image, and the presence or absence of the object in the grayscale image for determination is calculated from the calculation result Determining means for determining
When the shape of the object is substantially symmetric with respect to a symmetric element, the determination means obtains at least one of the template image and the gradation image for determination in the process of obtaining the image. An image processing apparatus that performs a correction process so as to form a gray image that is substantially symmetric with respect to the symmetric element, and performs a correlation value calculation using the gray image after correction.