WO2007013529A1 - 顔画像検出装置、顔画像検出方法および顔画像検出プログラム - Google Patents
顔画像検出装置、顔画像検出方法および顔画像検出プログラム Download PDFInfo
- Publication number
- WO2007013529A1 WO2007013529A1 PCT/JP2006/314806 JP2006314806W WO2007013529A1 WO 2007013529 A1 WO2007013529 A1 WO 2007013529A1 JP 2006314806 W JP2006314806 W JP 2006314806W WO 2007013529 A1 WO2007013529 A1 WO 2007013529A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- image
- face
- face part
- detected
- detection
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/165—Detection; Localisation; Normalisation using facial parts and geometric relationships
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
Definitions
- the present invention relates to a face image detection device, a face image detection method, and a face image detection program for detecting a face image present in an image, and in particular, improves detection accuracy of face images and performs detection processing.
- the present invention relates to a face image detection apparatus, a face image detection method, and a face image detection program that can reduce the time required for the operation.
- a face image detection technique for automatically recognizing whether or not a person's face appears in an image captured by a surveillance camera in a surveillance camera system or the like.
- face image detection techniques include the subspace method and the Integral Image method.
- Patent Document 1 Japanese Patent Application Laid-Open No. 2004-54442
- Patent Document 2 JP 2004-362468 A
- Non-Patent Literature 1 Paul Viola, Michael Jones, Rapid Object Detection using a Boosted Cascade of Simple Features ", In Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Volume 1, pp.511—518, December 20 01
- the area of the rectangular region for calculating the total value of feature amounts is relatively large in order to reduce the time required for the face image detection process. Is set. However, if the area of the rectangular area is large in an image where the direct sunlight hits the face, the total feature value may fluctuate greatly due to the influence of the direct sunlight, and face image detection may not be performed normally.
- the present invention has been made to solve the above-described problems caused by the prior art, and can improve the accuracy of detection of a face image and reduce the time required for detection processing.
- An object of the present invention is to provide a face image detection method and a face image detection program.
- a face image detection device is a face image detection device that detects a face image present in an image, and is a face part image.
- Extraction pattern storage means for storing information relating to a plurality of different extraction patterns for extracting a predetermined number of pixels from the peripheral pixels of the detection target pixel, and a plurality of different patterns stored in the extraction pattern storage means Extracting pixels by using the extraction pattern, and detecting the facial part image in the image using the extracted pixel feature amount, and the facial part image detected by the facial part image detecting unit And a face image detecting means for detecting a face image from the image.
- the face image detection apparatus further includes extraction pattern learning means for executing machine learning of the extraction pattern used for detection of the face part image in the invention of claim 1.
- the extraction pattern storage means stores information related to the extraction pattern machine-learned by the extraction pattern learning means.
- the face image detection apparatus is the case of the invention of claim 1 or 2, wherein the face part image detection means detects a face part image using a certain extraction pattern.
- the pixel that was the detection target when the face part image was not detected is excluded from the detection target pixel when the face part image is detected using the following extraction pattern. .
- the face image detection device is the face image detection device according to claim 1, 2 or 3, wherein the face part detection means uses the extracted feature amount of the pixel as a face image.
- a discriminant analysis for discriminating whether or not the image is a part image is executed, and a face part image is detected based on the result of the discriminant analysis.
- the face part image detection means is directed to a direction of an image pattern included in the image. And calculating the intensity and detecting the face part image in the image by using the calculated direction and intensity of the image pattern as the feature amount of the pixel.
- the face image detection means is a face detected by the face part image detection means. Based on the position of the part image, it is determined whether the image is a face image, and the face image is detected from the image based on the determination result.
- the face image detection device is the face image detection device according to the invention of claim 6, wherein the face image detection means is based on the position of the face part image detected by the face part detection means. Calculates the representative point of the image and determines whether the image is a face image based on the positional relationship between the calculated representative point and the position of the face part image detected by the face part image detecting means. It is characterized by doing.
- the face image detecting means divides the image into a plurality of regions on the basis of the representative point, and the face The facial image is detected by determining whether or not the image is a facial image based on information on a region where the facial part image is located when the facial part image is detected by the parts image detecting means. .
- the face image detection means uses information on a region where a pixel that is a detection target of a face part image is located. Then, discriminant analysis for discriminating whether or not the image is a face image is executed, and the face image is detected based on the result of the discriminant analysis.
- the face image detection method according to the invention of claim 10 is a face image detection method for detecting a face image existing in an image, and is a method for detecting peripheral pixels of a pixel to be detected from a face part image. Extracting pixels by using an extraction pattern storage step for storing information related to a plurality of different extraction patterns for extracting a predetermined number of pixels and a plurality of different extraction patterns stored in the extraction pattern storage step, A facial part image detection step of detecting a facial part image in the image using the extracted feature amount of the pixel, and the facial image obtained from the image power based on the facial part image detected by the facial part image detection step. And a face image detecting step of detecting.
- the face image detection program according to the invention of claim 11 is a face image detection program for detecting a face image existing in an image, and is a peripheral pixel of a pixel to be detected for a face part image.
- An extraction pattern storage procedure for storing information related to a plurality of different extraction patterns for extracting a predetermined number of pixels, and a plurality of different extraction patterns stored in accordance with the extraction pattern storage procedure are used to extract pixels.
- the facial part image detection procedure for detecting the facial part image in the image using the extracted feature amount of the pixel, and the facial image is detected from the image based on the facial part image detected by the facial part image detection procedure.
- information related to a plurality of different extraction patterns for extracting a predetermined number of pixels from the peripheral pixels of the pixel to be detected in the face part image is stored, and the stored different information Pixels are extracted by using multiple extraction patterns, and the feature values of the extracted pixels are used to detect facial part images in the image. Based on the detected facial part images, images are displayed. Image power Since face images are detected, the number of pixels to be extracted can be controlled to speed up the process of detecting face image images, thereby reducing the time required for face image detection. In addition, it is possible to improve the face detection accuracy by detecting the face part image, which is the basis for detecting the face image, a plurality of times.
- a discriminant analysis for discriminating whether or not the image is a face part image is executed using the extracted feature amount of the pixel, and based on the result of the discriminant analysis. Because it was decided to detect the face part image, it can be efficiently determined whether the image is a face part image, and the time required for the face image detection process can be further reduced. There is an effect.
- the direction and intensity of the image pattern included in the image are calculated, and the calculated image pattern direction and intensity are used as the feature amount of the pixel. Since the part image is detected, it is possible to accurately detect the image pattern such as the edge by using the direction and intensity of the image pattern as the feature amount, and to improve the detection accuracy of the face part image. .
- the image is a face image based on the position of the detected face part image, and the face image is converted from the image based on the determination result. Therefore, even if something wrong in the face part image is mistakenly detected as the face part image, it can be properly grasped and the detection accuracy of the face image can be improved. There is an effect.
- the representative point of the face image is calculated based on the position of the detected face part image, and the position between the calculated representative point and the detected face part image. Based on the relationship, it was decided whether or not the image is a face image. Therefore, the position of a point representing the face image in the image can be detected appropriately, and the detection accuracy of the face image can be improved. The effect is that it can be improved.
- the image is divided into a plurality of regions based on the representative point, and the image based on the information of the region where the face part image is located when the face part image is detected. Since it was decided to detect whether or not the face image was a face image, the face image was detected as a face part by examining which area the face part image was in. Even in this case, it can be detected appropriately and efficiently, and the effect of improving the detection accuracy of the face image is achieved.
- a discriminant analysis is performed to discriminate whether or not the image is a face image using the information of the area where the face part image is located, and the discriminant analysis is performed. Since the face image is detected based on the result, it is possible to efficiently determine whether or not the image is a face image, and to improve the detection accuracy of the face image. Play.
- information related to a plurality of different extraction patterns for extracting a predetermined number of pixels from the peripheral pixels of the pixel to be detected in the face part image is stored and stored. Extracting pixels by using multiple different extraction patterns, detecting the facial part image in the image using the extracted pixel features, and detecting the facial image from the image based on the detected facial part image Therefore, by controlling the number of pixels to be extracted, face part image detection processing can be performed at high speed, thereby reducing the time required for face image detection and reducing the face image. By detecting the face part image that is the basis for detecting an image a plurality of times, the face detection accuracy can be improved.
- information related to a plurality of different extraction patterns for extracting a predetermined number of pixels out of the peripheral pixels of the face part image detection target is stored and stored. Extracting pixels by using different extraction patterns, extracted pixels Since the face part image in the image is detected using the feature amount of the image and the face image is detected from the image based on the detected face part image, the number of pixels to be extracted can be controlled.
- the part image detection process can be performed at high speed, thereby reducing the time required to detect the face image and detecting the face part image that is the basis for detecting the face image multiple times. As a result, the face detection accuracy can be improved.
- FIG. 1 is a diagram for explaining the concept of face image detection processing according to the present invention.
- FIG. 2 is a diagram illustrating a functional configuration of the face image detection apparatus according to the present embodiment.
- FIG. 3 is a diagram for explaining a process of excluding detection target pixels of a face part image when a face part image is detected.
- FIG. 4 is a flowchart illustrating a processing procedure of face image detection processing according to the present embodiment.
- FIG. 5 is a flowchart showing a processing procedure of face part image detection processing shown in step S103 of FIG.
- FIG. 6 is a diagram showing a hardware configuration of the image processing apparatus shown in FIG.
- FIG. 7 is a diagram illustrating a determination process for determining whether or not there is a face part image at a position where a face part image is predicted to exist.
- FIG. 1 is a diagram for explaining the concept of face image detection processing according to the present invention.
- ⁇ : Lie is generated.
- the reduced images l la to l lc are generated.
- a plurality of enlarged images having different enlargement rates are generated.
- both the reduced image l la to l lc and the enlarged image are generated. This is because templates 12a to 12c described below can be applied to the image.
- a Gabor filter is applied to the pixel value of each pixel of the image, and the edge strength and the edge direction are calculated for each pixel. Then, rectangular templates 12a to 12c centering on the pixel to be detected of the face part image are selected, and the edge strength and the pixel intensity of the pixel to be detected and a predetermined number of pixels around the pixel are selected. And edge direction information is acquired.
- pixels corresponding to the four corner points of the rectangle and pixels corresponding to the midpoints of each side of the rectangle are extracted and detected as pixels around the pixel to be detected in the face part image. Together with the pixels to be extracted, information on edge strength and edge direction is obtained for a total of nine pixels.
- the templates 12a to 12c used have little lighting fluctuation Machine learning is performed so that the pixels of the face portion are selected.
- a discrimination score is calculated by inputting values of edge strength and edge direction into a linear discriminant that discriminates whether or not the image is a face part image, and based on the discriminant score value. The above determination is made.
- the shape of the templates 12a to 12c is not limited to the force quadrangle, but may be an ellipse or a circle. However, if the shape of the template 12a to 12c depends on the shape of the face part such as a specific person's eyes, the detection accuracy of other person's face parts may deteriorate, so the shape of the template 12a to 12c It is desirable to have a shape that does not depend on the shape of the facial parts.
- the number of pixels for acquiring edge strength and edge direction information is not limited to nine. Increasing the number of pixels increases the accuracy of facial part image detection, and reducing the number of pixels reduces the processing speed of facial part images. Therefore, by appropriately setting the number of pixels, the balance between the detection accuracy and the processing speed can be adjusted.
- a representative point 13 representing the position of the face image is determined based on the position of each detected face part image. Specifically, using the templates 12a to 12c, the direction in which the representative point 13 of the face image exists and the distance (number of pixels) are set in advance from the detection target pixel from which the face part image is detected. Keep it.
- the distribution of the position of the representative point 13 of the facial image is calculated from the position of each facial part image, and the representative point of the facial image is obtained by obtaining the peak of the distribution Determine 13.
- the image is divided by a predetermined number of pixels around the representative point 13 and divided into nine divided regions 14. Specifically, the range of each divided area 14 is set so that the four facial part images of the right eye, left eye, nose, and mouth are included in the upper left, upper right, center, and lower center divided areas 14, respectively.
- each divided region 14 when the image is a face image, the face part image should exist at a predetermined position and should not exist at other positions.
- the accuracy of face image detection can be improved.
- the total value of the discrimination scores of the four face part images of the right eye, left eye, nose, and mouth located in each divided region 14 is calculated, and whether the image is a face or a non-face is calculated from the total value.
- a 36-dimensional linear discriminant analysis for determining whether the image is a face image or a non-face image is performed using the generated feature amount. Specifically, a feature score is input to a linear discriminant to calculate a discriminant score, and when the discriminant score is larger than a predetermined threshold, it is determined that the image is a face image.
- FIG. 2 is a diagram illustrating a functional configuration of the face image detection apparatus according to the present embodiment.
- the face image detection apparatus includes an input unit 20, a display unit 21, a learning processing unit 22, a reduced / enlarged image generation unit 23, a Gabor feature image generation unit 24, a template selection unit 25, a face A part image detection unit 26, a face image representative point calculation unit 27, a face image detection feature amount generation unit 28, a face image detection processing unit 29, a storage unit 30, and a control unit 31 are included.
- the input unit 20 is an input device such as a keyboard or a mouse.
- the display unit 21 is a display device such as a display.
- the learning processing unit 22 is a processing unit that learns which template 12a to 12c is suitable for an image when detecting a face part image. Specifically, the learning processing unit 22 uses a boosting algorithm to learn which of the templates 12a to 12c can extract pixels and the like with little illumination variation.
- the reduced / enlarged image generation unit 23 receives the input image 10, a plurality of reduced images 1 la to which the input image 10 is reduced at different reduction ratios: 1 la to: It is a production
- whether the input image 10 is reduced or enlarged is determined by the relationship between the size of the templates 12 a to 12 c to be used and the predicted size of the face image included in the input image 10.
- the reduced / enlarged image generation unit 23 For example, if the size of the templates 12a to 12c to be used is about 20 pixels in length and width and the size of the face image to be predicted is 20 pixels in length and width, the reduced / enlarged image generation unit 23 generates the input image 10 An enlarged image of is generated. When the size of the face image cannot be predicted, the reduced / enlarged image generation unit 23 generates both reduced images lla to llc and an enlarged image.
- the Gabor feature image generation unit 24 is a generation unit that applies a Gabor filter to each image generated by the reduced / enlarged image generation unit 23 to generate a Gabor feature image. Specifically, the Gabor feature image generation unit 24 generates a Gabor feature image by selecting a 7-pixel square area from the image and applying Gabor filters to four directions every 90 degrees of the area. To do. Note that the size of 7 pixels square is set assuming that the right eye and left eye of the face image are about 12 pixels apart when the face image size is 24 pixels square. Have
- j is an imaginary unit
- 0 is a direction (angle)
- ⁇ is a wavelength
- ⁇ is a scalar
- k is four directions every 90 degrees.
- the Gabor feature image generation unit 24 calculates the values of the real part g real and the imaginary part g taag of the Gabor filter, and for each pixel in the value force image, the following five-dimensional element V ⁇
- a Gabor feature image is generated by calculating a feature amount that also has V force.
- the first dimension element V of the feature amount is
- the first dimension element V represents the strength of the edges in the image!
- the template selection unit 25 is a selection unit that selects a plurality of templates 12a to 12c suitable for an image using a result learned by the learning processing unit 22 when detecting a face part image.
- the face part image detection unit 26 is a detection unit that detects a face part image using the templates 12 a to 12 c selected by the template selection unit 25.
- the face part image detection unit 26 selects nine pixels of the image using the first templates 12a to 12c, and the edge intensity and edge direction of the Gabor feature image corresponding to these pixels. Based on this information, a linear discriminant analysis is performed to discriminate whether or not the image is a face part image.
- I is a variable to which the values of the five-dimensional feature values V to V at 9 points are assigned, and i is 1 to 45
- the coefficient a of the linear discriminant is preliminarily calculated so that it can be properly discriminated whether or not the image is a face part image.
- the five-dimensional features V to V at 9 points are preliminarily calculated so that it can be properly discriminated whether or not the image is a face part image.
- the face part image detection unit 26 calculates a value obtained by multiplying the weights of the used templates 12a to 12c and the discrimination score for each pixel that is the detection target of the face part image, and sets it as a collation value. If the collation value is smaller than the predetermined threshold value, it is determined that the image is not a face part image.
- the face part image detection unit 26 calculates a discrimination score using the following templates 12a to 12c, and uses the templates 12a to 12c used. A value obtained by multiplying the weight and the discriminant score is added to the collation value and set as a new collation value.
- the face part image detection unit 26 determines that the image is not a face part image! /.
- the face part image detection unit 26 determines that the image is not a face part image! /, And when the face part image is newly detected using the following templates 12a to 12c, the image is not displayed.
- the detection target pixels of the face part image when it is determined not to be a face part image are excluded from the detection target pixels when the face part image is detected using the following templates 12a to 12c.
- FIG. 3 is a diagram for explaining a process for removing a detection target pixel of a face part image when a face part image is detected.
- FIG. 3 shows an example in which the image of the right eye is detected, but the detection target pixel exclusion process is performed in the same manner when other face part images are detected.
- the face part image detection unit 26 sets the mask value of all pixels to “1” and sets all pixels to the right eye. It is assumed that the pixel to be detected when detecting this image.
- the face part image detection unit 26 applies the first templates 12a to 12c to detect the right eye image, and as a result, determines that the face part image detection unit 26 is not the right eye image.
- the value is set to “0”, and the processing is performed to exclude the pixel from the detection target pixel when the next template 12a to 12c is used to detect the image of the right eye.
- the facial image representative point calculation unit 27 detects the position of each facial part image detected by the facial part image detection unit 26.
- the position of the representative point 13 representing the position of the facial image.
- This is a calculation unit that calculates Specifically, the face image representative point calculation unit 27 determines in which direction the representative point 13 of the face image is in the direction of the detection target pixel when each face part image is detected using each of the templates 12a to 12c.
- the setting information related to the representative point that sets how many pixels the distance is is acquired.
- the face image representative point calculation unit 27 calculates the distribution of the positions of the representative points 13 of the face image based on the information related to the representative points and the information on the positions of the face part images.
- the position of the representative point 13 of the face image is determined by finding a point that is greater than or equal to the threshold value.
- the face image detection feature value generation unit 28 is a generation unit that generates a feature value used when detecting a face image from an image by linear discriminant analysis. More specifically, the face image detection feature quantity generation unit 28 divides the image by a predetermined number of pixels around the representative point 13 calculated by the face image representative point calculation unit 27, thereby dividing the image into nine divided regions. Divide into 14.
- the face image detection feature quantity generation unit 28 detects the total value of the collation values for each face part image obtained by applying the templates 12a to 12c, and detects the face part image. It is calculated for each divided area 14 where the target pixel is located, and 36 dimensions (9 areas X right eye, left eye, nose, mouth) for determining whether the image is a face image or a non-face image from the total value. ) Is generated.
- the face image detection processing unit 29 detects a face image with both image power by performing 36-dimensional linear discriminant analysis using the 36-dimensional feature amount generated by the face image detection feature amount generation unit 28. It is a processing unit.
- the linear discriminant used in this linear discriminant analysis is the same as equation (4). However, in this case, a and a are not enough to properly determine whether the image is a face image.
- w is a variable to which each value of 36-dimensional feature value is substituted, and i takes a value from 1 to 36.
- the storage unit 30 is a storage device such as a hard disk device.
- the storage unit 30 includes an input image 30a, a reduced and enlarged image 30b, a Gabor feature image 30c, template information 30d, a face part image detection linear discriminant 30e, a face part image discrimination result 30f, representative point setting information 30g, and a face image.
- Each data such as the detection linear discriminant 30h and the face image discrimination result 30i is stored.
- the input image 30a is an image for which a face image is detected.
- Reduced image 30b Are a plurality of images with different reduction ratios or enlargement ratios generated from the input image 30a.
- the reduced / enlarged image 30b corresponds to the reduced image lla to llc or the enlarged image described in FIG.
- the Gabor feature image 30c is an image having a 5D feature information power for each pixel obtained by applying a Gabor filter to the reduced and enlarged image 30b.
- the template information 30d is information on the templates 12a to 12c used when detecting the face part image. Specifically, the template information 30d is information on the relative positions of the eight pixels extracted when the face part image is detected with respect to the face part image detection target pixel, and information on the weights of the templates 12a to 12c. .
- the facial part image detection linear discriminant 30e is information on a linear discriminant used for detecting a face part image.
- the face part image discrimination result 30f stores information on the result of linear discriminant analysis when a face part image is detected. Specifically, the facial part image discrimination result 30f includes the position information of the detection target pixel that detected each face part image detected by the linear discriminant analysis, the information 12a to 12c of the template used, the discrimination score, and the collation value. Information is stored.
- the representative point setting information 30g stores the setting information of the positional relationship between each face part image and the representative point 13 of the face image. Specifically, the representative point setting information 30g includes information on the direction in which the representative point 13 exists and the distance (number of pixels) when the face part image is detected. Each face part is memorized.
- the face image detection linear discriminant 30h is information on a linear discriminant used when detecting a face image.
- the face image discrimination result 30i stores information on the result of linear discriminant analysis when a face image is detected. Specifically, the face image discrimination result 30i stores the position information of the representative point 13 of the face image, the discrimination score information when the image is a face image, and the like is discriminated by linear discriminant analysis. It is a thing.
- the control unit 31 is a control unit that controls the entire face image detection device, and controls data exchange between the functional units of the face image detection device.
- FIG. 4 is a flowchart illustrating the procedure of the face image detection process according to the present embodiment.
- the reduced / enlarged image generating unit 23 of the face image detecting device An input of an image to be detected is received (step S101), and a plurality of reduced images lla to 11c having different reduction ratios are generated from the received image (step S102).
- the reduced / enlarged image generation unit 23 generates reduced images lla to llc is described as an example, but the reduced / enlarged image generation unit 23 generates an enlarged image or the reduced image
- the same processing as described below is performed.
- the Gabor feature image generation unit 24, the template selection unit 25, and the face part image detection unit 26 are provided with a plurality of templates 12a for a plurality of reduced images l la to l lc having different reduction ratios.
- ⁇ 12c is applied, a collation value is calculated by linear discriminant analysis, and processing is performed to detect each face part image of the right eye, left eye, nose, and mouth (step S103). This face part image detection process will be described in detail with reference to FIG.
- the facial image representative point calculating unit 27 calculates the representative point of the facial image from the positions of the facial part images detected by the Gabor feature image generating unit 24, the template selecting unit 25, and the facial part image detecting unit 26.
- the position of 13 is calculated (step S104).
- the face image detection feature quantity generating unit 28 divides the image into nine parts by dividing the image by a predetermined number of pixels around the representative point 13 calculated by the face image representative point calculating unit 27.
- the image is divided into divided areas 14, and the total value of the matching values for each face part image is calculated for each divided area 14 where the detection target pixel that detected the face part image is located.
- a 36-dimensional feature value for determining whether the image is not a face image is generated (step S105).
- the face image detection processing unit 29 executes face image detection processing for detecting a face image from the image by linear discriminant analysis using 36-dimensional feature amounts (step S106), and the detection result of the face image Is output (step S107), and the face image detection process is terminated.
- FIG. 5 is a flowchart showing the procedure of the face part image detection process shown in step S103 of FIG.
- the Gabor feature image generation unit 24 generates a Gabor feature image in which each pixel is composed of a five-dimensional element using Equation (2) and Equation (3). (Step S201). Then, the face part image detection unit 26 first sets a face part for detecting an image. (Step S202).
- the face part image detection unit 26 initializes the mask value of all pixels of the image to “1” so that all pixels are detected pixels for detecting the face part image (Ste S203). Then, the template selection unit 25 selects templates 12a to 12c suitable for detecting the face part image from the input image 10 based on the result learned by the learning processing unit (step S204).
- the face part image detection unit 26 selects one target pixel for detecting the face part image (step S205), and checks whether or not the mask value of the selected pixel is “1” ( Step S).
- the face part image detection unit 26 uses the five-dimensional feature amount in the Gabor feature image of the pixel specified by the templates 12a to 12c.
- the linear discriminant analysis is performed (step S207), and the collation value used to determine whether the image is a face part image is calculated from the discriminant score and the weights of the templates 12a to 12c ( Step S208).
- the face part image detection unit 26 checks whether or not the collation value is greater than or equal to a predetermined threshold value.
- Step S209 If the collation value is equal to or greater than the predetermined threshold (Step S209, Yes), it is determined that the image is a face part image (Step S210), and the detection target pixel that detected the face part image is detected. Information such as the position, discrimination score, collation value, etc. (step S211
- the face part image detection unit 26 checks whether or not scanning of the image has been completed by selecting all the pixels (step S214). In step S209, if the collation value is greater than or equal to the predetermined threshold (step S209, No), the face part image detection unit 26 determines that the image is not a face part image (step S212), and the detection target.
- the mask value of the pixel is set to “0” so as to exclude the target pixel force for detecting the face part image from the pixel (step S213), and the process proceeds to step S214 as it is.
- step S206 If the mask value power of the selected pixel is not "l" in step S206 (step S206, No), the process proceeds to step S214 as it is.
- step S214 if the image scan is not yet completed (step S214, No), The image detection unit 26 selects the next detection target pixel for detecting the face part image (step S
- step S206 the process proceeds to step S206, and the subsequent processing is continued.
- the face part image detection unit 26 checks whether all the templates 12a to 12c have been used (step S216).
- the template selection unit 25 selects the next templates 12a to 12c (step S217).
- step S205 the process proceeds to step S205, and the subsequent processing is continued.
- the face part image detection unit 26 checks whether or not the detection processing of all the face part images is completed (Ste S218).
- step S218, No If the detection processing of all the facial part images has not been completed (step S218, No), the facial part image detection unit 26 does not set a facial part for detecting an image next. (Step S219), the process proceeds to step S203, and the subsequent processing is continued. If all the face part image detection processes have been completed (step S218, Yes), this face part image detection process is terminated.
- FIG. 6 is a diagram illustrating a hardware configuration of the image processing apparatus illustrated in FIG.
- this face image detection device includes a keyboard 100, a display 101, a ROM (Read Only Memory) 102, a medium reading device 103 that reads a program from a recording medium on which various programs are recorded, and other devices via a network.
- Network interface 104 that exchanges data with other computers, CPU (Central Processing Unit) 105, and HDD (Hard configuration).
- the HD (Hard Disk) 107 which is a storage medium that the HDD 106 stores and reads out, stores a face image detection program 107a that is realized by executing the face image detection method shown in the present embodiment on a computer. Then, after the face image detection program 107a is read out from the HD 107 at the time of execution, it is analyzed by the CPU 105, and the face image detection process 105a is executed.
- This face image detection process 105a includes the learning processing unit 22, the reduced and enlarged image generation unit 23, the Gabor feature image generation unit 24, the template selection unit 25, the face part image detection unit 26, and the face image shown in FIG. This corresponds to the functions of the representative point calculation unit 27, the face image detection feature value generation unit 28, and the face image detection processing unit 29.
- Various data 107b is stored in the HD 107.
- the various data 107b is stored in the RAM 108, and the various data 108a stored in the RAM 108 is referred to by the CPU 105.
- the various data 107b include the input image 30a, the reduced and enlarged image 30b, the gabor feature image 30c, the template information 30d, the face part image detection linear discriminant 30e, and the face part image discrimination result 30f shown in FIG. It corresponds to various data such as representative point setting information 30g, face image detection linear discriminant 30h, and face image discrimination result 30i.
- the storage unit 30 of the face image detection device extracts a plurality of different pixels from which a predetermined number of pixels are extracted from the peripheral pixels of the pixels to be detected as the face part image.
- Information on the templates 12a to 12c is stored, and the face part image detection unit 26 extracts pixels by using a plurality of different templates 12a to 12c stored in the storage unit 30, and uses the extracted feature values of the pixels.
- the face part image in the image is detected, and the face image detection unit 29 detects the face image based on the face part image detected by the face part detection unit 26.
- the facial part image detection process can be performed at high speed, thereby reducing the time required for the detection of the facial image and the facial part that is the basis for detecting the facial image. the image Face detection accuracy can be improved by detecting multiple times.
- the learning processing unit 22 executes machine learning of the templates 12a to 12c used for detection of the facial part images, and the storage unit 30 performs the template 12a on which machine learning has been performed. Since the information of ⁇ 12c is stored, it is possible to efficiently select an extraction pattern or the like that extracts pixels in a portion that is not easily affected by illumination fluctuations.
- the face part image detection unit 26 detects a face part image using a template 12a to 12c, the face part image is not detected and the powerful pixel is detected. Detection when detecting face part images using templates 12a to 12c Since it is excluded from the pixels to be output, the time required for the face image detection process can be further reduced.
- the face part image detection unit 26 performs a discriminant analysis to determine whether or not the image is a face part image using the pixel feature amounts extracted using the templates 12a to 12c. Since the facial part image is detected based on the result of the discriminant analysis, it is possible to efficiently determine whether the image is a facial part image. The time required can be further reduced.
- the Gabor feature image generation unit 24 calculates the direction and intensity of an image pattern such as an edge included in the input image 10, and the face part image detection unit 26 generates the Gabor feature image. Since the face part image in the image is detected by using the direction and intensity of the image pattern calculated by the unit 24 as the feature amount of the pixel, the edge is obtained by using the direction and intensity of the image pattern as the feature amount. Can be detected with high accuracy and the detection accuracy of facial part images can be improved.
- the face image detection unit 29 determines whether the image is a face image based on the position of the face part image detected by the face part image detection unit 26! Judgment and based on the result of the decision! / Image power Face image is detected, so even if it is a face part image, it is properly detected even if it is falsely detected as a face part image. And the detection accuracy of the face image can be improved.
- the face image representative point calculating unit 27 calculates the face representative point 13 based on the face part image detected by the face part image detecting unit 26, and the face image detecting unit 29 Whether or not the image is a face image based on the positional relationship between the representative point 13 calculated by the face image representative point calculating unit 27 and the face part image detected by the face part image detecting unit 26. Since the determination is made, the position of the point representing the face image in the image can be appropriately detected, and the detection accuracy of the face image can be improved.
- the face image detection feature quantity generation unit 28 divides the image into a plurality of divided regions 14 with the representative point 13 as a reference, and the face image detection unit 29 detects the face part image.
- the face image is detected by determining whether or not the image is a face image based on the information of the divided area 14 where the face part image is located. By examining which divided area 14 the pixel that was the target of output is, it can be detected properly even if a non-face part is mistakenly detected as a face part, improving the detection accuracy of the face image be able to.
- the face image detection unit 29 performs discriminant analysis for determining whether or not the image is a face image using information on the divided region 14 where the face part image is located, Since the face image is detected based on the result of the discrimination analysis, it can be efficiently determined whether or not the image is a face image, and the detection accuracy of the face image can be improved. it can.
- the image is divided into nine divided regions 14, and it is determined in which divided region 14 the face part image is detected, thereby determining whether the image is a face image.
- the present invention is not limited to this, and it is determined whether or not the image is a face image by determining whether or not there is a face part image at a position where a face part image is predicted to exist. As a matter of fact.
- FIG. 7 is a diagram for explaining a determination process for determining whether or not there is a face part image at a position where a face part image is predicted to exist. As shown in FIG. 7, in this determination process, the distribution (presence probability distributions 40a and 40b shown in FIG. 7) predicted to have a face part image based on the position of the representative point 13 of the face is preliminarily determined. Learn.
- the face image detection unit 29 determines that the face part is in an appropriate position when the position of the face part image is included in a range equal to or greater than the threshold in the distribution.
- the face image detection unit 29 performs this determination processing on all the face parts, and determines that the image is a face image when it is determined that all the face part images are in appropriate positions. .
- the face image detection unit 29 determines whether the image is a face image based on the positional relationship between the face part image and the representative point 13. By examining the positional relationship of the part image with respect to the representative point 13, even if a part that is not a face part image is erroneously detected as a face part image, it can be detected appropriately. The detection accuracy of the face image can be improved.
- the face part image or the face image is detected by performing linear discriminant analysis.
- the present invention is not limited to this, and the detection of the face part image or the face image is not limited to this. May be executed using other statistical methods. Other statistical methods include, for example, face part image or face image identification methods using nonlinear discriminant analysis, Support Vector Machine (SVM) method, neural network method, subspace method, and the like.
- SVM Support Vector Machine
- the constituent elements of the illustrated face image detection apparatus are functionally conceptual, and need not be physically configured as illustrated.
- the specific form of dispersion / integration of the face image detection device is not limited to the one shown in the figure, and all or a part thereof can be functionally or physically processed in an arbitrary unit according to various loads or usage conditions. Can be distributed and integrated.
- each or all of the processing functions performed in the face image detection device are realized by a CPU and a program that is analyzed and executed by the CPU, or hardware by wired logic. Can be realized as
- the face image detection method described in the present embodiment can be realized by executing a prepared program on a computer such as a personal computer or a workstation.
- This program can be distributed via a network such as the Internet.
- the program can also be executed by being recorded on a computer-readable recording medium such as a hard disk, a flexible disk (FD), a CD-ROM, an MO, and a DVD and being read by the computer.
- a computer-readable recording medium such as a hard disk, a flexible disk (FD), a CD-ROM, an MO, and a DVD and being read by the computer.
- the face image detection apparatus, the face image detection method, and the face image detection program according to the present invention need to improve the accuracy of face image detection and reduce the time required for face image detection processing. Useful for face image detection systems.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Geometry (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
Description
Claims
Priority Applications (6)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CA002616460A CA2616460A1 (en) | 2005-07-27 | 2006-07-26 | Face-image detecting apparatus, face-image detecting method, and face-image detecting program |
US11/996,569 US8311291B2 (en) | 2005-07-27 | 2006-07-26 | Face image detecting device, face image detecting method, and face image detecting program |
EP06781718.9A EP1909228B1 (en) | 2005-07-27 | 2006-07-26 | Face image detecting device, face image detecting method, and face image detecting program |
BRPI0614109A BRPI0614109A2 (pt) | 2005-07-27 | 2006-07-26 | aparelho de detecção de imagem facial que detecta uma imagem facial incluída em uma imagem, e, método e programa de detecção de imagem facial para detectar uma imagem facial incluída em uma imagem |
KR1020087002114A KR101445281B1 (ko) | 2005-07-27 | 2006-07-26 | 얼굴 화상 검출장치, 얼굴 화상 검출방법 및 얼굴 화상 검출프로그램을 기록한 컴퓨터 판독 가능한 기록매체 |
CN2006800269651A CN101228552B (zh) | 2005-07-27 | 2006-07-26 | 脸图像检测装置、脸图像检测方法 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2005-217711 | 2005-07-27 | ||
JP2005217711A JP4410732B2 (ja) | 2005-07-27 | 2005-07-27 | 顔画像検出装置、顔画像検出方法および顔画像検出プログラム |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2007013529A1 true WO2007013529A1 (ja) | 2007-02-01 |
Family
ID=37683420
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2006/314806 WO2007013529A1 (ja) | 2005-07-27 | 2006-07-26 | 顔画像検出装置、顔画像検出方法および顔画像検出プログラム |
Country Status (8)
Country | Link |
---|---|
US (1) | US8311291B2 (ja) |
EP (1) | EP1909228B1 (ja) |
JP (1) | JP4410732B2 (ja) |
KR (1) | KR101445281B1 (ja) |
CN (1) | CN101228552B (ja) |
BR (1) | BRPI0614109A2 (ja) |
CA (1) | CA2616460A1 (ja) |
WO (1) | WO2007013529A1 (ja) |
Families Citing this family (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP4532419B2 (ja) * | 2006-02-22 | 2010-08-25 | 富士フイルム株式会社 | 特徴点検出方法および装置並びにプログラム |
CN101315670B (zh) | 2007-06-01 | 2010-08-11 | 清华大学 | 特定被摄体检测装置及其学习装置和学习方法 |
JP2008311922A (ja) * | 2007-06-14 | 2008-12-25 | Fujifilm Corp | 撮像装置 |
JP2008311921A (ja) * | 2007-06-14 | 2008-12-25 | Fujifilm Corp | 撮像装置 |
EP2255950B1 (en) | 2007-08-09 | 2016-11-09 | Murata Machinery, Ltd. | Method for operating a filament winding apparatus |
JP5161311B2 (ja) * | 2007-09-19 | 2013-03-13 | トムソン ライセンシング | 画像をスケーリングするシステムおよび方法 |
JP4930433B2 (ja) * | 2008-04-01 | 2012-05-16 | セイコーエプソン株式会社 | 画像処理装置、画像処理方法、および画像処理プログラム |
JP4742193B2 (ja) * | 2009-04-28 | 2011-08-10 | Necソフト株式会社 | 年齢推定装置、年齢推定方法及びプログラム |
ES2377303B1 (es) * | 2009-06-05 | 2013-02-01 | Vodafone España S.A.U. | Método y sistema para recomendar fotografías. |
WO2011037579A1 (en) * | 2009-09-25 | 2011-03-31 | Hewlett-Packard Development Company, L.P. | Face recognition apparatus and methods |
JP2011170690A (ja) * | 2010-02-19 | 2011-09-01 | Sony Corp | 情報処理装置、情報処理方法、およびプログラム。 |
US9195501B2 (en) * | 2011-07-12 | 2015-11-24 | Qualcomm Incorporated | Instruction culling in graphics processing unit |
US8774508B2 (en) * | 2012-02-27 | 2014-07-08 | Denso It Laboratory, Inc. | Local feature amount calculating device, method of calculating local feature amount, corresponding point searching apparatus, and method of searching corresponding point |
WO2013138268A1 (en) * | 2012-03-12 | 2013-09-19 | Diy, Co. | Automatic face detection and parental approval in images and video and applications thereof |
US9443137B2 (en) * | 2012-05-08 | 2016-09-13 | Samsung Electronics Co., Ltd. | Apparatus and method for detecting body parts |
CN103455234A (zh) * | 2012-06-01 | 2013-12-18 | 腾讯科技(深圳)有限公司 | 显示应用程序界面的方法及装置 |
EP3054677A4 (en) * | 2013-09-30 | 2017-05-10 | Coolpad Software Tech (Shenzhen) Co., Ltd. | Methods and systems for image encoding and decoding and terminal |
WO2015049826A1 (ja) * | 2013-10-01 | 2015-04-09 | 日本電気株式会社 | 物体検出装置、物体検出方法および学習装置 |
US9807316B2 (en) * | 2014-09-04 | 2017-10-31 | Htc Corporation | Method for image segmentation |
WO2020049933A1 (ja) * | 2018-09-05 | 2020-03-12 | 日本電産株式会社 | 物体認識装置および物体認識方法 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH07220090A (ja) * | 1994-02-02 | 1995-08-18 | Canon Inc | 物体認識方法 |
JPH11120351A (ja) * | 1997-10-15 | 1999-04-30 | Fujitsu Ltd | 画像マッチング装置および画像マッチングプログラムを格納する記憶媒体 |
JP2003317084A (ja) * | 2002-04-19 | 2003-11-07 | Nec Corp | 顔画像からの目検出システム、目検出方法および目検出用プログラム |
JP2004054442A (ja) * | 2002-07-17 | 2004-02-19 | Glory Ltd | 顔検出装置、顔検出方法および顔検出プログラム |
Family Cites Families (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6463176B1 (en) | 1994-02-02 | 2002-10-08 | Canon Kabushiki Kaisha | Image recognition/reproduction method and apparatus |
US5710590A (en) * | 1994-04-15 | 1998-01-20 | Hitachi, Ltd. | Image signal encoding and communicating apparatus using means for extracting particular portions of an object image |
US6108437A (en) * | 1997-11-14 | 2000-08-22 | Seiko Epson Corporation | Face recognition apparatus, method, system and computer readable medium thereof |
GB2341231A (en) * | 1998-09-05 | 2000-03-08 | Sharp Kk | Face detection in an image |
JP3636927B2 (ja) * | 1999-05-18 | 2005-04-06 | 三菱電機株式会社 | 顔画像処理装置 |
EP1107166A3 (en) * | 1999-12-01 | 2008-08-06 | Matsushita Electric Industrial Co., Ltd. | Device and method for face image extraction, and recording medium having recorded program for the method |
JP2001216515A (ja) * | 2000-02-01 | 2001-08-10 | Matsushita Electric Ind Co Ltd | 人物の顔の検出方法およびその装置 |
JP2001285787A (ja) | 2000-03-31 | 2001-10-12 | Nec Corp | 映像録画方法およびそのシステムとその記録媒体 |
US7155036B2 (en) * | 2000-12-04 | 2006-12-26 | Sony Corporation | Face detection under varying rotation |
JP3846851B2 (ja) | 2001-02-01 | 2006-11-15 | 松下電器産業株式会社 | 画像のマッチング処理方法及びその装置 |
GB0112773D0 (en) * | 2001-05-25 | 2001-07-18 | Univ Manchester | Object identification |
US7050607B2 (en) * | 2001-12-08 | 2006-05-23 | Microsoft Corp. | System and method for multi-view face detection |
AUPS140502A0 (en) * | 2002-03-27 | 2002-05-09 | Seeing Machines Pty Ltd | Method for automatic detection of facial features |
JP4447245B2 (ja) | 2003-06-06 | 2010-04-07 | オムロン株式会社 | 特定被写体検出装置 |
US7783082B2 (en) * | 2003-06-30 | 2010-08-24 | Honda Motor Co., Ltd. | System and method for face recognition |
US7920725B2 (en) * | 2003-09-09 | 2011-04-05 | Fujifilm Corporation | Apparatus, method, and program for discriminating subjects |
-
2005
- 2005-07-27 JP JP2005217711A patent/JP4410732B2/ja not_active Expired - Fee Related
-
2006
- 2006-07-26 KR KR1020087002114A patent/KR101445281B1/ko active IP Right Grant
- 2006-07-26 EP EP06781718.9A patent/EP1909228B1/en not_active Expired - Fee Related
- 2006-07-26 CN CN2006800269651A patent/CN101228552B/zh not_active Expired - Fee Related
- 2006-07-26 CA CA002616460A patent/CA2616460A1/en not_active Abandoned
- 2006-07-26 BR BRPI0614109A patent/BRPI0614109A2/pt not_active IP Right Cessation
- 2006-07-26 WO PCT/JP2006/314806 patent/WO2007013529A1/ja active Application Filing
- 2006-07-26 US US11/996,569 patent/US8311291B2/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH07220090A (ja) * | 1994-02-02 | 1995-08-18 | Canon Inc | 物体認識方法 |
JPH11120351A (ja) * | 1997-10-15 | 1999-04-30 | Fujitsu Ltd | 画像マッチング装置および画像マッチングプログラムを格納する記憶媒体 |
JP2003317084A (ja) * | 2002-04-19 | 2003-11-07 | Nec Corp | 顔画像からの目検出システム、目検出方法および目検出用プログラム |
JP2004054442A (ja) * | 2002-07-17 | 2004-02-19 | Glory Ltd | 顔検出装置、顔検出方法および顔検出プログラム |
Also Published As
Publication number | Publication date |
---|---|
CN101228552A (zh) | 2008-07-23 |
EP1909228A1 (en) | 2008-04-09 |
JP4410732B2 (ja) | 2010-02-03 |
EP1909228B1 (en) | 2014-07-02 |
KR20080031031A (ko) | 2008-04-07 |
CA2616460A1 (en) | 2007-02-01 |
KR101445281B1 (ko) | 2014-09-26 |
BRPI0614109A2 (pt) | 2016-11-22 |
CN101228552B (zh) | 2012-10-03 |
US20090041357A1 (en) | 2009-02-12 |
JP2007034723A (ja) | 2007-02-08 |
EP1909228A4 (en) | 2009-07-01 |
US8311291B2 (en) | 2012-11-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP4410732B2 (ja) | 顔画像検出装置、顔画像検出方法および顔画像検出プログラム | |
US7840037B2 (en) | Adaptive scanning for performance enhancement in image detection systems | |
Xu et al. | Automatic 3D face recognition from depth and intensity Gabor features | |
Everingham et al. | Regression and classification approaches to eye localization in face images | |
US8401250B2 (en) | Detecting objects of interest in still images | |
EP1918850A2 (en) | Method and apparatus for detecting faces in digital images | |
US7844085B2 (en) | Pairwise feature learning with boosting for use in face detection | |
KR20080033486A (ko) | 서포트 벡터 머신 및 얼굴 인식에 기초한 자동 생체 식별 | |
JP2011198268A (ja) | 情報処理装置および方法、並びにプログラム | |
Levinshtein et al. | Hybrid eye center localization using cascaded regression and hand-crafted model fitting | |
Zakaria et al. | Hierarchical skin-adaboost-neural network (h-skann) for multi-face detection | |
JP2007025900A (ja) | 画像処理装置、画像処理方法 | |
JP2006323779A (ja) | 画像処理方法、画像処理装置 | |
Zhang et al. | Hierarchical facial landmark localization via cascaded random binary patterns | |
Mahdi et al. | 3D facial matching by spiral convolutional metric learning and a biometric fusion-net of demographic properties | |
Goswami et al. | Kernel group sparse representation based classifier for multimodal biometrics | |
JP2006285959A (ja) | 顔判別装置の学習方法、顔判別方法および装置並びにプログラム | |
Xu et al. | Facial expression pervasive analysis based on haar-like features and svm | |
JP2006244385A (ja) | 顔判別装置およびプログラム並びに顔判別装置の学習方法 | |
Azmi et al. | Freeman chain code (FCC) representation in signature fraud detection based on nearest neighbour and artificial neural network (ANN) classifiers | |
Ganakwar et al. | Comparative analysis of various face detection methods | |
El-Abed et al. | A SVM-based model for the evaluation of biometric sample quality | |
CN111832460B (zh) | 一种基于多特征融合的人脸图像提取方法及系统 | |
Huang et al. | Statistical learning of evaluation function for ASM/AAM image alignment | |
Sánchez López | Local Binary Patterns applied to Face Detection and Recognition |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 200680026965.1 Country of ref document: CN |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
WWE | Wipo information: entry into national phase |
Ref document number: 11996569 Country of ref document: US Ref document number: 2006781718 Country of ref document: EP |
|
ENP | Entry into the national phase |
Ref document number: 2616460 Country of ref document: CA |
|
WWE | Wipo information: entry into national phase |
Ref document number: 1020087002114 Country of ref document: KR |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: PI0614109 Country of ref document: BR Kind code of ref document: A2 Effective date: 20080125 |