• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    In a broadly-applicable face image extraction device and method for defining a face by position and size in target images varied in type for face image extraction at high speed, an edge extraction part 1 extracts an edge part from a target image and generates an edge image. A template storage part 2 previously stores a template composed of a plurality of concentric shapes varied in size. A voting result storage part 3 has voting storage regions for each size of the concentric shapes of the template so as to store the result obtained by voting processing carried out by a voting part 4. The voting part 4 carries out the voting processing utilizing the template at each pixel in the edge image, and stores the result obtained thereby in the corresponding voting storage region. After the voting processing, an analysis part 5 performs cluster evaluation based on the voting results stored in the voting storage regions, and then defines the face in the target image by position and size.
    公开号:US20010002932A1
    申请号:US09/725,751
    申请日:2000-11-30
    公开日:2001-06-07
    发明作者:Hideaki Matsuo；Kazuyuki Imagawa；Yuji Takata；Naruatsu Baba；Toshiaki Ejima
    申请人:Panasonic Corp；
    IPC主号:G06K9-00234
    专利说明:
    [0001] 1. Field of the Invention [0001]
    [0002] The present invention relates to a device and a method for face image extraction, and a recording medium having recorded a program for carrying out the method. More specifically, in image processing, such device and method are used to extract, at high speed, a face region from a target image utilizing a template to define position and size thereof. [0002]
    [0003] 2. Description of the Background Art [0003]
    [0004] As everyone acknowledges, a human face often mirrors his/her thinking and feeling, and thus is considered a significant factor. In image processing especially where handling human images, if such human face can be automatically detected and processed to reveal its position and size in a target image, such system comes in useful. Here, the target image includes still pictures and moving pictures, and a person taken therein may be both real and artificial created by computer graphics, for example. This is the reason for the recent attempt in image processing to extract a face region out of any target image on such system. [0004]
    [0005] Conventional technologies of such face image extraction have been disclosed in Japanese Patent Laid-Open Publication No. 9-73544 (97-73544) (hereinafter, first document) and No. 10-307923 (98-307923) (hereinafter, second document), for example. [0005]
    [0006] The technology disclosed in the first document is of finding an approximation of face region by an ellipse. Therein, the ellipse is defined by five parameters including center coordinates (x, y), a radius r, a ratio b between major and minor axes, and an angle θ between the major axis and an x axis. These parameters are changed as appropriate to be optimal in value for face image extraction. [0006]
    [0007] In the second document, the technology is of successively finding face parts (e.g., eyes, nose, mouth). [0007]
    [0008] In the first document, however, approximation requires repeated calculation to change those parameters (especially the angle θ takes time). In consideration of a face image hardly staying the same, real-time approximation is hopeless with the processing capability of existing personal computers, so thus is real-time face image extraction processing. Also in this technology, there has no concern given for a possibility that one image may include several human faces, and thus applicability of this technology is considered narrow. [0008]
    [0009] In the second document, the technology is not available unless otherwise a face region has been defined by position in an image. Therefore, this is applicable only to a specific image, resulting in narrow applicability. [0009] SUMMARY OF THE INVENTION
    [0010] Therefore, an object of the present invention is to provide a broadly-applicable device and method for defining a face by position and size in images varied in type for face image extraction at high speed, and a recording medium having recorded a program for carrying out the method. [0010]
    [0011] The present invention has the following features to attain the object above. [0011]
    [0012] A first aspect of the present invention is directed to a face image extraction device for defining a face in a target image by position and size for extraction, comprising: [0012]
    [0013] an edge extraction part for extracting an edge part (pixels outlining a person or face) from the target image, and generating an image having only the edge part (hereinafter, edge image); [0013]
    [0014] a template storage part for storing a template composed of a plurality of predetermined concentric shapes equal in shape but varied in size; [0014]
    [0015] a voting result storage part for storing, in a interrelating manner, voting values and coordinates of pixels on the edge image for every size of the concentric shapes of the template; [0015]
    [0016] a voting part for increasing or decreasing the voting values of every pixel, specified by the coordinates, outlining each of the concentric shapes every time a center point of the template moves on the pixels in the edge part; and [0016]
    [0017] an analysis part for defining the face in the target image by position and size based on the voting values stored in the voting result storage part. [0017]
    [0018] As described above, in the first aspect, the face position can be detected at high speed only with light-loaded voting processing and evaluation of voting values. Further, as is utilizing a template composed of concentric shapes varied in size, approximation can be done in a practical manner by comparing, in size, an edge part presumed to include a face region with the template. Accordingly, size of the face can be detected also at high speed. As such, in the face image extraction device of the present invention, processing load can be considerably reduced, thereby achieving almost real-time face region extraction even with the processing capabilities available for the existing personal computers. Further, in the first aspect, a face region does not have to be defined where and how many in a target image prior to extraction, and thus a face can be detected no matter what size and type the target image is. Accordingly, applicability of the device considered quite wide. [0018]
    [0019] Herein, preferably, the predetermined concentric shape is a circle, an ellipse, or a polygon. In such case, the circle may improve the voting result in accuracy as is being constant in distance from a center point to each pixel outlining the circle. [0019]
    [0020] Preferably, the edge extraction part extracts the edge part from the target image by using a filter for a high frequency component. [0020]
    [0021] Therefore, any high frequency component can be obtained by using a filter for the target image, whereby position and size of a face can be preferably detected in a case where the target image is a still picture. [0021]
    [0022] Preferably, when the target image is structured by a plurality of successive images, the edge extraction part extracts the edge part by comparing a current image with another image temporally before, and with after to calculate a difference therebetween, respectively, for every image structuring the target image. [0022]
    [0023] In this manner, a current target image is compared with another temporally before and then with after to calculate a difference therebetween, respectively. Accordingly, position and size of a face can be preferably detected in a case where the target image is a series of moving pictures. Further, with the help of a template for detection, a face region can be stably extracted at high-speed even if facial expression changes to a greater extent at zoom-in or close-up, for example. [0023]
    [0024] Also preferably, the edge extraction part detects, with respect to pixels extracted in every predetermined box, one pixel located far-left end or far-right end in the box on a scanning line basis, and regards only the pixels detected thereby as the edge part. [0024]
    [0025] In this manner, any part differed in texture within contour is prevented from being extracted as the edge part. Therefore, the extraction processing can be done, at high speed, with respect to the face region. [0025]
    [0026] Also preferably, the analysis part performs clustering with respect to the voting values stored in each of the voting result storage parts, and narrows down position and size of the face in the target image. [0026]
    [0027] Therefore, even in the case that a target image includes several faces, the face region can be extracted by clustering the voting results (each voting value) and then correctly evaluating correlation thereamong. [0027]
    [0028] Also preferably, the face image extraction device further comprises an image editing part for editing the target image in a predetermined manner by distinguishing a face region defined by position and size in the analysis part from the rest in the target image. [0028]
    [0029] As such, by editing the target image while distinguishing a face region defined by position and size from the rest, only a desired part, i.e., a face, can be emphasized and thus become conspicuous in the target image As an example, the target image excluding the face region may be solidly shaded, leading to eye-catching effects. [0029]
    [0030] Still preferably, the face image extraction device further comprises an image editing part for replacing an image of the face region defined by position and size by the analysis part with another. [0030]
    [0031] As such, the image of the face region can be replaced with another. In this manner, the face can be intentionally concealed. This works effective, for example, when image-monitoring a person who is suffering dementia. In such case, by replacing the image of a face with another, privacy can be protected, and a face area can be defined for monitoring This works also good when replacing images of a person's movement with other type of character's. [0031]
    [0032] A second aspect of the present invention is directed to a face image extraction method for defining a face in a target image by position and size for extraction, comprising: [0032]
    [0033] an extraction step of extracting an edge part (pixels outlining a person or face) from the target image, and generating an image having only the edge part (hereinafter, edge image); [0033]
    [0034] a first storage step of storing a template composed of a plurality of predetermined concentric shapes equal in shape but varied in size; [0034]
    [0035] a second storage step of storing, in a interrelating manner, voting values and coordinates of pixels on the edge image for every size of the concentric shapes of the template; [0035]
    [0036] a voting step of increasing or decreasing the voting values of every pixel, specified by the coordinates, outlining each of the concentric shapes every time a center point of the template moves on the pixels in the edge part; and [0036]
    [0037] an analysis step of defining, after the voting step, the face in the target image by position and size based on the voting values. [0037]
    [0038] As described above, in the second aspect, the face position can be detected at high speed only with light-loaded voting processing and evaluation of voting values. Further, as is utilizing a template composed of concentric shapes varied in size, approximation can be done in a practical manner by comparing, in size, an edge part presumed to include a face region with the template. Accordingly, size of the face can be detected also at high speed. As such, in the face image extraction device of the present invention, processing load can be considerably reduced, thereby achieving almost real-time face region extraction even with the processing capabilities available for the existing personal computers. Further, in the second aspect, a face region does not have to be defined where and how many in a target image prior to extraction, and thus a face can be detected no matter what size and type the target image is. Accordingly, applicability of the device considered quite wide. [0038]
    [0039] Herein, preferably, the predetermined concentric shape is a circle, an ellipse, or a polygon. [0039]
    [0040] In such case, the circle may improve the voting result in accuracy as is being constant in distance from a center point to each pixel outlining the circle. [0040]
    [0041] Also preferably, in the extraction step, the edge part is extracted from the target image by using a filter for a high frequency component. [0041]
    [0042] Accordingly, a high frequency component is extracted from the target image by using a filter. Therefore, position and size of a face can be preferably detected in a case where the target image is a still picture. [0042]
    [0043] Also preferably, when the target image is structured by a plurality of successive images, the edge part is extracted by comparing a current image with another image temporally before, and with after to calculate a difference therebetween, respectively, for every image structuring the target image. [0043]
    [0044] In this manner, a current target image is compared with another temporally before and then with after to calculate a difference therebetween, respectively. Accordingly, position and size of a face can be preferably detected in a case where the target image is a series of moving pictures. Further, with the help of a template for detection, a face region can be stably extracted at high-speed even if facial expression changes to a greater extent at zoom-in or close-up, for example. [0044]
    [0045] Also preferably, in the extraction step, with respect to pixels extracted in every predetermined box, one pixel located far-left end or far-right end in the box is detected on a scanning line basis, and only the pixels detected thereby is regarded as the edge part. [0045]
    [0046] As such, any part differed in texture within contour is prevented from being extracted as the edge part. Therefore, the extraction processing can be done, at high speed, with respect to the face region. [0046]
    [0047] Still preferably, in the analysis step, clustering is performed with respect to the voting values stored in each of the voting result storage parts, and position and size of the face is narrowed down in the target image. [0047]
    [0048] As such, even in the case that a target image includes several faces, the face region can be extracted by clustering the voting results (each voting value) and then correctly evaluating correlation thereamong. [0048]
    [0049] These and other objects, features, aspects and advantages of the present invention will become more apparent from the following detailed description of the present invention when taken in conjunction with the accompanying drawings. [0049] BRIEF DESCRIPTION OF THE DRAWINGS
    [0050] FIG. 1 is a block diagram showing the structure of a face image extraction device according to one embodiment of the present invention; [0050]
    [0051] FIGS. 2[0051] a and 2 b are diagrams each showing an exemplary structure of an edge extraction part 1;
    [0052] FIGS. 3[0052] a to 3 c are diagrams each showing an exemplary edge image extracted by the edge extraction part 1;
    [0053] FIGS. 4[0053] a to 4 c are diagrams each showing an exemplary template stored in a template storage part 2;
    [0054] FIG. 5 is a flowchart showing the procedure of voting processing carried out by a voting part [0054] 4;
    [0055] FIG. 6 is a diagram in assistance of explaining the concept of voting values stored, through voting processing, in voting storage regions provided in a voting result storage part [0055] 3;
    [0056] FIG. 7 is a flowchart showing the procedure of analysis processing carried out by an analysis part [0056] 5;
    [0057] FIGS. 8[0057] a to 8 c are diagrams in assistance of explaining the concept of clustering processing carried out in steps S23 and S24 in FIG. 7; and
    [0058] FIGS. 9[0058] a to 9 c are diagrams showing an exemplary image edit processing carried out by an image editing part 6. DESCRIPTION OF THE PREFERRED EMBODIMENTS
    [0059] FIG. 1 is a block diagram showing the structure of a face image extraction device according to an embodiment of the present invention. In FIG. 1, the face image extraction device of the embodiment includes an edge extraction part [0059] 1, a template storage part 2, a voting result storage part 3, a voting part 4, an analysis part 5, and an image editing part 6.
    [0060] Referring to the accompanying drawings, described below is the operation of each component above and a method for face image extraction. [0060]
    [0061] The edge extraction part [0061] 1 receives an image for face image extraction (hereinafter, target image), and extracts an edge part therefrom to generate another image having only the edge part (hereinafter, edge image). Here, the edge part is a part (pixels) representing contours of human body or face, for example, where high in frequency. The target image may be both still and moving, and depending on which, a technique applied for edge part extraction differs.
    [0062] For a still picture, as shown in FIG. 2[0062] a, the edge extraction part 1 is implemented by a filter 11 which takes out only a high frequency component, thereby simplifying edge part extraction process. The preferable type of the filter 11 is a Sobel.
    [0063] For moving pictures, as shown in FIG. 2[0063] b, the edge extraction part 1 is implemented by a difference extraction part 12. Specifically, the difference extraction part 12 compares a targeted moving picture with another located temporally before and then with after to calculate a difference therebetween (data difference on pixel basis), respectively. Thereafter, any part found large in such detected difference (where motion in images is active) is extracted as an edge part.
    [0064] Here, with the above techniques, a part(s) differed in texture within contours is extracted also as the edge part. FIG. 3[0064] a shows an exemplary edge image including such unwanted extracted parts. Although this causes no problem for the face image extraction device of the present invention, the following technique is preferable if processing therein is desired to be faster.
    [0065] First, in such edge image as shown in FIG. 3[0065] a, any area having edge part rather concentrated is enclosed in a box (FIG. 3b). The image in the box is then subjected to bi-directional scanning on scanning line basis (FIG. 3b), and any outline formed by pixels each detected first thereby is determined as being the edge part in the target image (FIG. 3c). In this manner, any part differed in texture within contour is prevented from being extracted. Any constituent for this processing may be provided subsequent to the filter 11 or the difference extraction part 12.
    [0066] The template storage part [0066] 2 previously stores data about a template composed of a plurality of concentric shapes, which are equal in shape but varied in size. Here, although the concentric shape may be any such as circle, ellipse, regular polygon, or polygon, but most preferable is circle. This is because the distance from a center point to an outline of the shape. i.e., to each pixel outlining the circle, is always constant, thereby improving the later-described voting result in accuracy.
    [0067] Here, as shown in FIGS. 4[0067] a to 4 c, a template described in this embodiment is presumed to be composed of concentric circles t1 to tn (where n is an arbitrary integer) each differed in radius from a center point P. As for those circles t1 to tn, the difference in radius may be uniform as is a template T1 of FIG. 4a or irregular as is a template T2 of FIG. 4b. Further, those circles of t1 to tn may be outlined by one-dot line (correspond to a pixel in a target image) as is the template T2 of FIG. 4b, or as a template T3 of FIG. 4c, some or all of those may be outlined by two-dot or thicker line (i.e., annular ring shape). Hereinafter, a term “circle” means both circle and annular ring.
    [0068] The circles t[0068] 1 to tn are stored in the template storage part 2 as one template, but practically, handled as each independent. Therefore, for each of the circles t1 to tn, pixel data is stored in the template storage part 2 in the form of table, for example.
    [0069] The voting result storage part [0069] 3 has regions each for the shapes of the template stored in the template storage part 2, in this example, circles t1 to tn. The regions (hereinafter, referred to as voting storage regions) store a result obtained by voting processing carried out by the voting part 4, which will be described later. Herein, the number of voting storage regions provided in the voting result storage part 3 is equal to that of circles, in this example, n. Note herein that, each voting storage region is of the same size as a target image.
    [0070] As for the edge image generated by the edge extraction part [0070] 1, the voting part 4 carries out the voting processing utilizing the template stored in the template storage part 2. FIG. 5 is a flowchart showing the procedure of the voting processing.
    [0071] Referring to FIG. 5, the voting part [0071] 4 first accesses the voting result storage part 3, and initializes, to 0, components (voting values) representing x-y coordinates in each voting storage region (step S11). Thereafter, the voting part 4 sets the center point P of the template at the head of pixels in the edge part on the edge image (step S12). To find the head of pixels, the edge image is sequentially scanned, vertically or laterally, from the upper left. The position of pixel found first in the edge part may be regarded as the head.
    [0072] The voting part [0072] 4 then initializes, to “1”, a counter i indicates which of the shapes of the template (in this example, circles t1 to tn) (step S13). When the counter i indicates 1, for example, the voting part 4 uses the circle t1 and specifies every component outlining the circle t1 on the edge image by x-y coordinates (step S14). The voting part 4 then adds, i.e., votes, “1” to each of the components specified by the x-y coordinates in the voting storage region provided for the circle t1 in the voting result storage part 3. This is the voting processing.
    [0073] Thereafter, the voting part [0073] 4 increments the counter i, i=2 (step S17). Since the counter i is now indicating the circle t2, the voting part 4 then specifies every component outlining the circle t2 by x-y coordinates (step S14). The voting part 4 then adds “1” again to each of the components specified by the x-y coordinates in the voting storage region this time provided for the circle t2 in the voting result storage part 3 (step S15).
    [0074] As for the circles t[0074] 3 to tn, the voting part 4 repeats the voting processing in steps S14 and S15 in the same manner as above while incrementing the counter i until i becomes n (steps S16, S17). As such, the voting storage regions provided each for the circles t1 to tn store the voting result obtained through the voting processing carried out at the head pixel in the edge part.
    [0075] Thereafter, the voting part [0075] 4 sets the center point P of the template at a pixel next to the head pixel, and then repeats the processing in steps S13 to S17. This is done for every pixel, a pixel at a time, in the edge part on the edge image (steps S18, S19). In short, the center point P of the template never misses a single pixel in the edge part for the voting processing.
    [0076] As an example, by subjecting the above-described voting processing to such edge image as shown in FIG. 3[0076] c, n voting storage regions provided in the voting result storage 3 store such voting values as shown in FIG. 6. Here, presumably, the edge image shown in FIG. 3c is subjected to the above voting processing. For the sake of simplicity, FIG. 6 shows a case where the voting processing is carried out only at a specific pixel in the edge part. In each of the voting storage regions of FIG. 6, a circle is outlined by the components representing x-y coordinates having the voting value of “1”. Here, since the voting value is accumulated as described in the foregoing, a part where the circles in FIG. 6 are crossing (indicated by a dot) has the larger voting value.
    [0077] Accordingly, if the above-described voting processing is done to pixels being an edge part representing contours of a face approximated by a circle or an ellipse, the voting value is found larger in the vicinity of a center point thereof. It means that any part found larger in voting value is highly-possible to be the center of the face. Such phenomenon of the voting value concentrating at a specific part becomes noticeable when the concentric shape is a circle having a radius equal to or almost equal to a minimum width of the edge part. In consideration thereof, by determining in which voting storage region such phenomenon is conspicuously observed, the face can be specified by size. This sounds similar to generalized Hough transformation. However, the face image extraction method of the present invention is absolutely different therefrom in a respect that a face region can be specified, simultaneously, by position and size. This is implemented by using a template composed of concentric shapes varied in size. [0077]
    [0078] Here, in the voting processing, the components representing x-y coordinates in each voting storage region may be initialized to a predetermined maximum value in step S[0078] 11, and then the voting part 4 may subtract “1” from each applicable component in step S15. If this is the case, any part found smaller in voting value is highly-possible to be the center of the face, and by determining in which voting storage region such phenomenon is conspicuously observed, the face can be specified by size.
    [0079] Moreover, in step S[0079] 15, the value adding to or subtracting from the voting value is not restricted to “1”, and may be arbitrarily determined.
    [0080] Described next is a technique for specifying a face region in a target image according to the voting results stored in the voting result storage part [0080] 3.
    [0081] Once the voting part [0081] 4 completed its voting processing, the analysis part 5 refers to the voting results stored in the voting result storage part 3 for cluster evaluation, and then specifies a face in the target image by position and size. FIG. 7 is a flowchart showing the procedure of analysis processing carried out by the analysis part 5.
    [0082] Referring to FIG. 7, the analysis part [0082] 5 first sets a counter j, to “1”, whose value indicates which of the shapes of the template (in this example, circles t1 to tn) (step S21). When the counter j indicates 1, for example, the analysis part 5 refers to the voting storage region corresponding to the circle t1 for the voting result stored therein. The analysis part 5 then extracts any component whose voting value is exceeding a predetermined value of G (e.g., 200) (step S22). This threshold value G can be arbitrarily determined based on definition of the target image and a desired accuracy for face image extraction. The analysis part 5 performs clustering only for the extracted component(s) (step S23), and as for each clustered region, calculates variance and covariance values (step S24). In order to determine similarlity among clustered regions, any of Euclidean squared distance, generalized Euclidean squared distance, Maharanobis distance, or Minkowski distance may be applied. Further, to form clusters, any of SLINK (single linkage clustering method), CLINK (complete linkage clustering method) or UPGMA (unweighted pair-group method using arithmetic averages) may be applied.
    [0083] The analysis part [0083] 5 then compares the variance and covariance values calculated for each clustered region with a predetermined threshold value of H (step S25). If those values are found smaller than the threshold value H in step S25, the analysis part 5 regards a center point in the face region as a center point of the face. Assuming that the counter j indicates “1”, the size (diameter) of the circle t1 is determined as being a minor axis in length (step S26), and a length obtained by adding a constant (empirically determined) to the minor axis is as a major axis of the face (step S27). The analysis part 5 stores thus determined center point, and minor and major axes as the analysis results (step S28). On the other hand, if the variance and covariance values are found equal to or larger than the threshold value H, the analysis part 5 determines the center point in the region is not a center point of the face, and then the procedure moves to the next processing.
    [0084] Thereafter, the analysis part [0084] 5 increments the counter j, j=2 (step S30). Since the counter j is now indicating the circle t2, the analysis part 5 then refers to the voting result stored in the voting storage region corresponding to the circle t2, and then extracts any component whose voting value is exceeding the threshold value G (step S22). The analysis part 5 performs clustering only for the extracted component(s) (step S23), and as for each clustered region, calculates variance and covariance values (step S24).
    [0085] The analysis part [0085] 5 then compares the variance and covariance values calculated for each clustered region with a predetermined threshold value of H (step S25). If those values are found smaller than the threshold value H in step S25, the analysis part 5 regards a center point in the face region as a center point of the face. Assuming that the counter j indicates “1”, the size of the circle t2 is determined as being a minor axis in length (step S26), and a length obtained by adding a constant (empirically determined) to the minor axis is as a major axis of the face (step S27). The analysis part 5 additionally stores thus determined center point, and minor and major axes as the analysis results (step S28). On the other hand, if the variance and covariance values are found equal to or larger than the threshold value H, the analysis part 5 determines the center point in the region is not a center point of the face, and then the procedure moves to the next processing.
    [0086] As for the circles t[0086] 3 to tn, the analysis part 5 repeats the analysis processing in steps S22 to S28 in the same manner as above while incrementing the counter j until j becomes n (steps S29, S30). As such, stored are the analysis results obtained through the analysis processing carried out for face image extraction for the voting storage regions provided each for the circles t1 to tn.
    [0087] The analysis results are then outputted to the image editing part [0087] 6.
    [0088] Here, with reference to FIGS. 8[0088] a to 8 c, the clustering carried out in steps S23 and S24 is briefly described.
    [0089] Assuming herein is that a case where components exceeding the threshold value G in voting value (dots in the drawings) are distributed as in FIG. 8[0089] a. Cluster evaluation performed in such case by the analysis part 5 is as follows. In the initial clustering, exemplarily as in FIG. 8b, four initial clusters of A, B, C, and D are generated. Once those initial clusters were generated, then similarity is calculated for every pair of clusters. If the calculated similarity is equal to or larger than a predetermined threshold value, the applicable pair of clusters are combined. In FIG. 8c, exemplarily, the clusters C and D are combined, and becomes a cluster E. Thereafter, the clusters A, B, and E are calculated for a variance value, and the like, for evaluation. Herein, since the cluster A and B are both small in variance value, center points thereof are both considered a center of the face. The cluster E large in variance value is not considered a center of the face.
    [0090] In the case that two or more clusters are detected by evaluation made based on the variance value, for example, determination of a face region may be done as follows: [0090]
    [0091] First, if the detected clusters share a center point and varied in size, a face region is the cluster whose variance value is minimum; [0091]
    [0092] Second, if the detected clusters do not share a center point and varied in size, all of those are face regions each differed in position and size; and [0092]
    [0093] Third, if the detected clusters do not share a center point but identical in size, all of those are face regions differed in position but same in size. [0093]
    [0094] The image editing part [0094] 6 receives the analysis results (face regions) from the analysis part 5, and responds to any request for image processing with respect to the target image. Utilized herein is a face region being distinguishable from the rest by the analysis results. For example, the image editing part 6 clips or solidly shades the target image of FIG. 9a, leaving only a face region (FIG. 9b). Accordingly, generated thereby is an image having only a face emphasized. Alternatively, the image of the face region of FIG. 9a can be replaced with another (e.g., image of other character's face) as shown in FIG. 9c. In this manner, the face can be intentionally concealed.
    [0095] Note that, the image editing part [0095] 6 is appropriately provided to meet a need for image processing utilizing the extracted face region, but is not essential for the face image extraction device of the present invention.
    [0096] As is known from the above, according to the face image extraction device and method of the present embodiment, face position can be detected at high speed only with light-loaded voting processing (basically, only addition) and evaluation of voting values. Further, as is utilizing a template composed of concentric shapes varied in size, approximation can be done in a practical manner by comparing, in size, an edge part presumed to be a face region with the template. Accordingly, size of the face can be detected also at high speed. As such, in the face image extraction device of the present invention, processing load can be considerably reduced, thereby achieving almost real-time face region extraction even with the processing capabilities available for the existing personal computers. [0096]
    [0097] Further, with the face image extraction device of the present invention, a face region does not have to be defined where and how many in a target image prior to extraction, and thus a face can be detected no matter what size and type the target image is. Accordingly, applicability of the device is considered quite wide. Moreover, even in the case that the target image includes several faces, the face region can be extracted by clustering the voting results and then correctly evaluating correlation thereamong. [0097]
    [0098] Typically, the face image extraction device of the above embodiment is functionally (face image extraction method) implemented by a storage having a predetermined program stored therein (e.g., ROM, RAM, hard disk) and a CPU (Central Processing Unit) carrying out the program. Here, the program may be provided by a recording medium such as CD-ROM or floppy disk. The program may be partially recorded in a plurality of recording media for distribution. [0098]
    [0099] It is herein assumed that a part of the program is functionally put on various processes or threads (e.g., DLL) no matter whether the program being a part of operating system or not. In such case, even if not storing the part of the program, the recording medium is regarded as the one having recorded the program for carrying out the face image extraction method of the present invention. [0099]
    [0100] Moreover, described in the foregoing is the exemplary case that the face image extraction method of the present invention is implemented by a stand-alone type (FIG. 1), but this is not restrictive and may be implemented by a server-client type. In other words, in addition to the stand-alone type having only one terminal functionally carry out the face image extraction method, the server-client type will also do. Therein, the face image extraction method is partially or entirely carried out functionally by a server or a device on a network connectable to a terminal being a client. For example, the server may be the one functionally carrying out the method, and the client has only a WWW browser. In such case, information (e.g., template, voting values) is normally on the server, and is distributed to the client basically over the network. When the information is on the server, a storage in the server is equivalent to the “recording medium”, and when on the client, a storage in the client is equivalent thereto. [0100]
    [0101] Further, the program carrying out the face image extraction method of the present invention may be an application written in machine language after compilation, or an intermediate code interpreted by the above process or thread. Or, in a “recording medium”, at least resource and source codes are stored together with a compiler and a linker, which can generate an application written in machine language by utilizing such codes. Or, in a “recording medium”, at least the resource and source codes are stored together with an interpreter, which can generate an application in the intermediate code by utilizing such codes. [0101]
    [0102] While the invention has been described in detail, the foregoing description is in all aspects illustrative and not restrictive. It is understood that numerous other modifications and variations can be devised without departing from the scope of the invention. [0102]
    权利要求:
    Claims (19)
    [1" id="US-20010002932-A1-CLM-00001] 1. A face image extraction device for defining a face in a target image by position and size for extraction, comprising:
    an edge extraction part for extracting an edge part (pixels outlining a person or face) from said target image, and generating an image having only the edge part (hereinafter, edge image);
    a template storage part for storing a template composed of a plurality of predetermined concentric shapes equal in shape but varied in size;
    a voting result storage part for storing, in a interrelating manner, voting values and coordinates of pixels on said edge image for every size of the concentric shapes of said template;
    a voting part for increasing or decreasing said voting values of every pixel, specified by the coordinates, outlining each of said concentric shapes every time a center point of said template moves on the pixels in said edge part; and
    an analysis part for defining the face in said target image by position and size based on said voting values stored in said voting result storage part.
    [2" id="US-20010002932-A1-CLM-00002] 2. The face image extraction device according to
    claim 1 , wherein said predetermined concentric shape is a circle.
    [3" id="US-20010002932-A1-CLM-00003] 3. The face image extraction device according to
    claim 1 , wherein said predetermined concentric shape is an ellipse.
    [4" id="US-20010002932-A1-CLM-00004] 4. The face image extraction device according to
    claim 1 , wherein said predetermined concentric shape is a polygon.
    [5" id="US-20010002932-A1-CLM-00005] 5. The face image extraction device according to
    claim 1 , wherein said edge extraction part extracts said edge part from said target image by using a filter for a high frequency component.
    [6" id="US-20010002932-A1-CLM-00006] 6. The face image extraction device according to
    claim 1 , wherein, when said target image is structured by a plurality of successive images, said edge extraction part extracts said edge part by comparing a current image with another image temporally before, and with after to calculate a difference therebetween, respectively, for every image structuring said target image.
    [7" id="US-20010002932-A1-CLM-00007] 7. The face image extraction device according to
    claim 1 , wherein said edge extraction part detects, with respect to pixels extracted in every predetermined box, one pixel located far-left end or far-right end in the box on a scanning line basis, and regards only the pixels detected thereby as said edge part.
    [8" id="US-20010002932-A1-CLM-00008] 8. The face image extraction device according to
    claim 1 , wherein said analysis part performs clustering with respect to said voting values stored in each of said voting result storage parts, and narrows down position and size of the face in said target image.
    [9" id="US-20010002932-A1-CLM-00009] 9. The face image extraction device according to
    claim 1 , further comprising an image editing part for editing said target image in a predetermined manner by distinguishing a face region defined by position and size in said analysis part from the rest in the target image.
    [10" id="US-20010002932-A1-CLM-00010] 10. The face image extraction device according to
    claim 1 , further comprising an image editing part for replacing an image of the face region defined by position and size by said analysis part with another.
    [11" id="US-20010002932-A1-CLM-00011] 11. A face image extraction method for defining a face in a target image by position and size for extraction, comprising:
    an extraction step of extracting an edge part (pixels outlining a person or face) from said target image, and generating an image having only the edge part (hereinafter, edge image);
    a first storage step of storing a template composed of a plurality of predetermined concentric shapes equal in shape but varied in size;
    a second storage step of storing, in a interrelating manner, voting values and coordinates of pixels on said edge image for every size of the concentric shapes of said template;
    a voting step of increasing or decreasing said voting values of every pixel, specified by the coordinates, outlining each of said concentric shapes every time a center point of said template moves on the pixels in said edge part; and
    an analysis step of defining, after said voting step, the face in said target image by position and size based on said voting values.
    [12" id="US-20010002932-A1-CLM-00012] 12. The face image extraction method according to
    claim 11 , wherein said predetermined concentric shape is a circle.
    [13" id="US-20010002932-A1-CLM-00013] 13. The face image extraction method according to
    claim 11 , wherein said predetermined concentric shape is an ellipse.
    [14" id="US-20010002932-A1-CLM-00014] 14. The face image extraction method according to
    claim 11 , wherein said predetermined concentric shape is a polygon.
    [15" id="US-20010002932-A1-CLM-00015] 15. The face image extraction device according to
    claim 11 , wherein, in said extraction step, said edge part is extracted from said target image by using a filter for a high frequency component.
    [16" id="US-20010002932-A1-CLM-00016] 16. The face image extraction method according to
    claim 11 , wherein, in said extraction step, when said target image is structured by a plurality of successive images, said edge part is extracted by comparing a current image with another image temporally before, and with after to calculate a difference therebetween, respectively, for every image structuring said target image.
    [17" id="US-20010002932-A1-CLM-00017] 17. The face image extraction method according to
    claim 11 , wherein, in said extraction step, with respect to pixels extracted in every predetermined box, one pixel located far-left end or far-right end in the box is detected on a scanning line basis, and only the pixels detected thereby is regarded as said edge part.
    [18" id="US-20010002932-A1-CLM-00018] 18. The face image extraction method according to
    claim 11 , wherein, in said analysis step, clustering is performed with respect to said voting values stored in each of said voting result storage parts, and position and size of the face is narrowed down in said target image.
    [19" id="US-20010002932-A1-CLM-00019] 19. A recording medium having recorded a face image extraction method for defining a face in a target image by position and size as a program executable on a computer device, the program at least comprising:
    a extraction step of extracting an edge part (pixels outlining a person or face) from said target image, and generating an image having only the edge part (hereinafter, edge image);
    a first storage step of storing a template composed of a plurality of predetermined concentric shapes equal in shape but varied in size;
    a second storage part of storing, in a interrelating manner, voting values and coordinates of pixels on said edge image for every size of the concentric shapes of said template;
    a voting step of increasing or decreasing said voting values of every pixel outlining each of said concentric shapes every time a center point of said template moves on the pixel in said edge part; and
    an analysis step of defining, after said voting step, the face in said target image by position and size based on said voting values.
    类似技术:
    公开号 | 公开日 | 专利标题
    US6697503B2|2004-02-24|Device and method for face image extraction, and recording medium having recorded program for the method
    JP3218004B2|2001-10-15|Test signature verification method
    Aleksic et al.2006|Automatic facial expression recognition using facial animation parameters and multistream HMMs
    US5715325A|1998-02-03|Apparatus and method for detecting a face in a video image
    US7266225B2|2007-09-04|Face direction estimation using a single gray-level image
    Bhattacharya et al.2013|Offline signature verification using pixel matching technique
    WO2001033497A1|2001-05-10|A system and method for face detection through geometric distribution of a non-intensity image property
    EP1271394A2|2003-01-02|Method for automatically locating eyes in an image
    Chan et al.2011|Local ordinal contrast pattern histograms for spatiotemporal, lip-based speaker authentication
    KR20010103631A|2001-11-23|System and method for biometrics-based facial feature extraction
    US6915022B2|2005-07-05|Image preprocessing method capable of increasing the accuracy of face detection
    Jida et al.2017|Face segmentation and detection using Voronoi diagram and 2D histogram
    KR100696251B1|2007-03-20|Method and apparatus for setting of comparison area and generating of user authentication information for iris recognition
    Joosten et al.2015|Voice activity detection based on facial movement
    Wakasugi et al.2004|Robust lip contour extraction using separability of multi-dimensional distributions
    Hotta2003|View-invariant face detection method based on local pca cells
    Bharadi et al.2018|Multi-modal biometric recognition using human iris and dynamic pressure variation of handwritten signatures
    JP2001222719A|2001-08-17|Face extracting device, face extracting method and recording medium for face extraction program
    WO2009144330A1|2009-12-03|Method for detection of objectionable contents in still or animated digital images
    US20030059117A1|2003-03-27|Systems and methods for image processing, and recording medium therefor
    Vélez et al.2013|Robust ear detection for biometric verification
    JP2002008042A|2002-01-11|Action recognizing and processing device, and moving object action analyzing system
    GB2329739A|1999-03-31|Fuzzy-neural face recognition
    Sehad et al.2000|Face recognition under varying views
    JPH08153187A|1996-06-11|Image recognizing method
    同族专利:
    公开号 | 公开日
    EP1107166A3|2008-08-06|
    US6697503B2|2004-02-24|
    EP1107166A2|2001-06-13|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US5828769A|1996-10-23|1998-10-27|Autodesk, Inc.|Method and apparatus for recognition of objects via position and orientation consensus of local image encoding|US20030059117A1|2001-09-27|2003-03-27|Matsushita Electric Industrial Co., Ltd.|Systems and methods for image processing, and recording medium therefor|
    US20040032985A1|2002-07-12|2004-02-19|Minolta Co., Ltd.|Edge image acquisition apparatus capable of accurately extracting an edge of a moving object|
    US20050128307A1|2003-12-10|2005-06-16|Sony Corporation|Image processing method and apparatus and, program|
    US20050232487A1|2004-04-14|2005-10-20|Safeview, Inc.|Active subject privacy imaging|
    US20050265605A1|2004-05-28|2005-12-01|Eiji Nakamoto|Object recognition system|
    JP2006508463A|2002-11-29|2006-03-09|ソニー・ユナイテッド・キングダム・リミテッド|Face detection|
    US20060104480A1|2004-11-12|2006-05-18|Safeview, Inc.|Active subject imaging with body identification|
    US20090041357A1|2005-07-27|2009-02-12|Toru Yonezawa|Face image detecting device, face image detecting method, and face image detecting program|
    US20100277636A1|2005-02-07|2010-11-04|Panasonic Corporation|Imaging device|
    US8041081B2|2006-06-28|2011-10-18|Fujifilm Corporation|Method, apparatus, and program for human figure region extraction|
    US20120025997A1|2010-05-27|2012-02-02|University Of Southern California|System and method for failure prediction for rod pump artificial lift systems|
    US20130324244A1|2012-06-04|2013-12-05|Sony Computer Entertainment Inc.|Managing controller pairing in a multiplayer game|
    US20140153817A1|2012-11-30|2014-06-05|Adobe Systems Incorporated|Patch Size Adaptation for Image Enhancement|
    US8988237B2|2010-05-27|2015-03-24|University Of Southern California|System and method for failure prediction for artificial lift systems|
    US9117262B2|2012-11-30|2015-08-25|Adobe Systems Incorporated|Learned piece-wise patch regression for image enhancement|
    US9157308B2|2011-12-29|2015-10-13|Chevron U.S.A. Inc.|System and method for prioritizing artificial lift system failure alerts|
    CN107633535A|2017-09-06|2018-01-26|深圳市易天自动化设备股份有限公司|A kind of high fast positioning method of new machine sensation target|
    US20190052819A1|2017-11-29|2019-02-14|Intel Corporation|Methods, apparatus and articles of manufacture to protect sensitive information in video collaboration systems|JP3797686B2|1995-09-07|2006-07-19|株式会社東芝|Shape recognition apparatus and method|
    US6184926B1|1996-11-26|2001-02-06|Ncr Corporation|System and method for detecting a human face in uncontrolled environments|
    JPH10307923A|1997-05-01|1998-11-17|Mitsubishi Electric Corp|Face parts extraction device and face direction detection device|US8330831B2|2003-08-05|2012-12-11|DigitalOptics Corporation Europe Limited|Method of gathering visual meta data using a reference image|
    US7092569B1|1999-07-29|2006-08-15|Fuji Photo Film Co., Ltd.|Method and device for extracting specified image subjects|
    CN1186936C|2000-05-22|2005-01-26|松下电器产业株式会社|Image communication terminal|
    DE10043460C2|2000-09-04|2003-01-30|Fraunhofer Ges Forschung|Locating parts of the body by evaluating edge direction information|
    JP3784289B2|2000-09-12|2006-06-07|松下電器産業株式会社|Media editing method and apparatus|
    US20030007700A1|2001-07-03|2003-01-09|Koninklijke Philips Electronics N.V.|Method and apparatus for interleaving a user image in an original image sequence|
    WO2003010728A1|2001-07-24|2003-02-06|Koninklijke Kpn N.V.|Method and system and data source for processing of image data|
    EP1280117A1|2001-07-24|2003-01-29|Koninklijke KPN N.V.|Method and system for processing of image data|
    EP1308916A1|2001-11-05|2003-05-07|Koninklijke KPN N.V.|Method and system and data source for processing of image data|
    DE10158990C1|2001-11-30|2003-04-10|Bosch Gmbh Robert|Video surveillance system incorporates masking of identified object for maintaining privacy until entry of authorisation|
    US6959099B2|2001-12-06|2005-10-25|Koninklijke Philips Electronics N.V.|Method and apparatus for automatic face blurring|
    EP1357515A1|2002-04-22|2003-10-29|Agfa-Gevaert AG|Method for processing digital image data of photographic images|
    US8498452B2|2003-06-26|2013-07-30|DigitalOptics Corporation Europe Limited|Digital image processing using face detection information|
    US7565030B2|2003-06-26|2009-07-21|Fotonation Vision Limited|Detecting orientation of digital images using face detection information|
    US7574016B2|2003-06-26|2009-08-11|Fotonation Vision Limited|Digital image processing using face detection information|
    US7440593B1|2003-06-26|2008-10-21|Fotonation Vision Limited|Method of improving orientation and color balance of digital images using face detection information|
    US7616233B2|2003-06-26|2009-11-10|Fotonation Vision Limited|Perfecting of digital image capture parameters within acquisition devices using face detection|
    US7471846B2|2003-06-26|2008-12-30|Fotonation Vision Limited|Perfecting the effect of flash within an image acquisition devices using face detection|
    US8989453B2|2003-06-26|2015-03-24|Fotonation Limited|Digital image processing using face detection information|
    US9692964B2|2003-06-26|2017-06-27|Fotonation Limited|Modification of post-viewing parameters for digital images using image region or feature information|
    US7362368B2|2003-06-26|2008-04-22|Fotonation Vision Limited|Perfecting the optics within a digital image acquisition device using face detection|
    US7844076B2|2003-06-26|2010-11-30|Fotonation Vision Limited|Digital image processing using face detection and skin tone information|
    US9129381B2|2003-06-26|2015-09-08|Fotonation Limited|Modification of post-viewing parameters for digital images using image region or feature information|
    US7269292B2|2003-06-26|2007-09-11|Fotonation Vision Limited|Digital image adjustable compression and resolution using face detection information|
    US8948468B2|2003-06-26|2015-02-03|Fotonation Limited|Modification of viewing parameters for digital images using face detection information|
    JP2005122351A|2003-10-15|2005-05-12|Seiko Epson Corp|Method, system and program for searching for face image candidate area|
    US8553949B2|2004-01-22|2013-10-08|DigitalOptics Corporation Europe Limited|Classification and organization of consumer digital images using workflow, and face detection and recognition|
    US7564994B1|2004-01-22|2009-07-21|Fotonation Vision Limited|Classification system for consumer digital images using automatic workflow and face detection and recognition|
    GB2412831A|2004-03-30|2005-10-05|Univ Newcastle|Highlighting important information by blurring less important information|
    JP4449576B2|2004-05-28|2010-04-14|パナソニック電工株式会社|Image processing method and image processing apparatus|
    US8320641B2|2004-10-28|2012-11-27|DigitalOptics Corporation Europe Limited|Method and apparatus for red-eye detection using preview or other reference images|
    US7715597B2|2004-12-29|2010-05-11|Fotonation Ireland Limited|Method and component for image recognition|
    JP4975272B2|2005-05-30|2012-07-11|京セラ株式会社|User terminal|
    US7792970B2|2005-06-17|2010-09-07|Fotonation Vision Limited|Method for establishing a paired connection between media devices|
    US8593542B2|2005-12-27|2013-11-26|DigitalOptics Corporation Europe Limited|Foreground/background separation using reference images|
    US8682097B2|2006-02-14|2014-03-25|DigitalOptics Corporation Europe Limited|Digital image enhancement with reference images|
    US7804983B2|2006-02-24|2010-09-28|Fotonation Vision Limited|Digital image acquisition control and correction method and apparatus|
    US7792335B2|2006-02-24|2010-09-07|Fotonation Vision Limited|Method and apparatus for selective disqualification of digital images|
    DE602007012246D1|2006-06-12|2011-03-10|Tessera Tech Ireland Ltd|PROGRESS IN EXTENDING THE AAM TECHNIQUES FROM GRAY CALENDAR TO COLOR PICTURES|
    US7515740B2|2006-08-02|2009-04-07|Fotonation Vision Limited|Face recognition with combined PCA-based datasets|
    US7916897B2|2006-08-11|2011-03-29|Tessera Technologies Ireland Limited|Face tracking for controlling imaging parameters|
    US7403643B2|2006-08-11|2008-07-22|Fotonation Vision Limited|Real-time face tracking in a digital image acquisition device|
    US7620218B2|2006-08-11|2009-11-17|Fotonation Ireland Limited|Real-time face tracking with reference images|
    US7315631B1|2006-08-11|2008-01-01|Fotonation Vision Limited|Real-time face tracking in a digital image acquisition device|
    US8055067B2|2007-01-18|2011-11-08|DigitalOptics Corporation Europe Limited|Color segmentation|
    WO2008104549A2|2007-02-28|2008-09-04|Fotonation Vision Limited|Separating directional lighting variability in statistical face modelling based on texture space decomposition|
    US8503800B2|2007-03-05|2013-08-06|DigitalOptics Corporation Europe Limited|Illumination detection using classifier chains|
    KR101247147B1|2007-03-05|2013-03-29|디지털옵틱스 코포레이션 유럽 리미티드|Face searching and detection in a digital image acquisition device|
    US8189927B2|2007-03-05|2012-05-29|DigitalOptics Corporation Europe Limited|Face categorization and annotation of a mobile phone contact list|
    US8363951B2|2007-03-05|2013-01-29|DigitalOptics Corporation Europe Limited|Face recognition training method and apparatus|
    US7916971B2|2007-05-24|2011-03-29|Tessera Technologies Ireland Limited|Image processing method and apparatus|
    US8896725B2|2007-06-21|2014-11-25|Fotonation Limited|Image capture device with contemporaneous reference image capture mechanism|
    US8155397B2|2007-09-26|2012-04-10|DigitalOptics Corporation Europe Limited|Face tracking in a camera processor|
    US8705810B2|2007-12-28|2014-04-22|Intel Corporation|Detecting and indexing characters of videos by NCuts and page ranking|
    US8750578B2|2008-01-29|2014-06-10|DigitalOptics Corporation Europe Limited|Detecting facial expressions in digital images|
    US8494286B2|2008-02-05|2013-07-23|DigitalOptics Corporation Europe Limited|Face detection in mid-shot digital images|
    US7855737B2|2008-03-26|2010-12-21|Fotonation Ireland Limited|Method of making a digital camera image of a scene including the camera user|
    JP5547730B2|2008-07-30|2014-07-16|デジタルオプティックス・コーポレイション・ヨーロッパ・リミテッド|Automatic facial and skin beautification using face detection|
    WO2010063463A2|2008-12-05|2010-06-10|Fotonation Ireland Limited|Face recognition using face tracker classifier data|
    US8488023B2|2009-05-20|2013-07-16|DigitalOptics Corporation Europe Limited|Identifying facial expressions in acquired digital images|
    US8379917B2|2009-10-02|2013-02-19|DigitalOptics Corporation Europe Limited|Face recognition performance using additional image features|
    法律状态:
    2000-11-30| AS| Assignment|Owner name: MATSUSHITA ELECTRIC INDUSTRIAL CO., LTD., JAPAN Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:MATSUO, HIDEAKI;IMAGAWA, KAZUYUKI;TAKATA, YUJI;AND OTHERS;REEL/FRAME:011323/0470 Effective date: 20000929 |
    2004-02-05| STCF| Information on status: patent grant|Free format text: PATENTED CASE |
    2007-07-27| FPAY| Fee payment|Year of fee payment: 4 |
    2011-07-27| FPAY| Fee payment|Year of fee payment: 8 |
    2014-05-27| AS| Assignment|Owner name: PANASONIC INTELLECTUAL PROPERTY CORPORATION OF AMERICA, CALIFORNIA Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:PANASONIC CORPORATION;REEL/FRAME:033033/0163 Effective date: 20140527 Owner name: PANASONIC INTELLECTUAL PROPERTY CORPORATION OF AME Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:PANASONIC CORPORATION;REEL/FRAME:033033/0163 Effective date: 20140527 |
    2015-07-15| FPAY| Fee payment|Year of fee payment: 12 |
    优先权:
    申请号 | 申请日 | 专利标题
    JP342025/1999||1999-12-01||
    JP11-342025||1999-12-01||
    JP34202599||1999-12-01||
    [返回顶部]