JP2006048328A - Apparatus and method for detecting face - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide face detection technology capable of simply and accurately detecting a face area also in an image to which reflected light or the like from spectacles is reflected. <P>SOLUTION: In the face detection apparatus, an image is scanned by the window WD of a fixed size and whether a face is present in the window WD or not is decided to detect a face. When a person in the image wears spectacles and external light or the like is reflected to the spectacles, eye parts are shielded and the accuracy of face detection is reduced. Thereby a high luminance part HB is detected from an image part in the window WD and the detected high luminance part HB is converted into a low luminance part LB corresponding to black, for example. Consequently face detection can be simply and accurately performed even in an image to which the reflected light or the like from the spectacles is reflected. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a face detection technique for detecting a face area from an image.

  A technique for detecting a face area from an image is disclosed in Patent Document 1, for example. In this face detection technique, an image to be detected is scanned in a window of a predetermined size to determine whether a face is included in the window. For this determination, after the pixels in the window are vectorized and projected onto the principal component space, the distance to the principal component space of the vector and the distance in the principal component space are calculated, and the face area exists in the window. It is determined whether or not to do so.

  As another face detection technique, for example, Patent Document 2 discloses a technique using a background difference method. This technology specifies the positions of the face area and the eye area from the difference information between a previously acquired background image (an image in which no person is shown) and an image in which a person is shown. When the eye portion is blocked by reflection of external light or the like because it is applied, this portion is corrected from surrounding pixel information.

JP-A-8-3339445 JP 2002-269545 A

  However, with the technique of Patent Document 1, it is difficult to detect a face with high accuracy when the face existing in the window is not in an appropriate state. For example, when detecting the face of a person wearing glasses, the accuracy of face detection will be significantly reduced if the eyes are shielded by external light reflected on the glasses.

  On the other hand, in the technique of Patent Document 2, since the background subtraction method is used, face detection is impossible for an image for which a background image cannot be obtained in the first place, and face detection cannot be easily performed.

  The present invention has been made in view of the above-described problems, and an object of the present invention is to provide a face detection technique that can easily and accurately detect a face region even for an image in which reflected light from a pair of glasses is reflected.

  In order to solve the above-mentioned problem, the invention of claim 1 is a face detection device for detecting a face region from an image, wherein (a) an acquisition means for acquiring an image, and (b) acquired by the acquisition means With respect to a detection area of a certain size included in the image, by changing the position of the detection area in the image, scanning means for scanning the image, and (c) each of the detection areas changed by the scanning means An image part corresponding to the detection area is extracted from the image at a position, and a face detection unit that detects a face area of the image part is provided.The face detection unit includes (c-1) the image part. A high luminance detection means for detecting a high luminance part; and (c-2) a removing means for removing the high luminance part from the image part when the high luminance part is detected by the high luminance detection means. Have.

  The invention of claim 2 is a face detection device for detecting a face area from an image, wherein (a) an acquisition means for acquiring an original image, and (b) a resolution of the original image acquired by the acquisition means. Resolution conversion means for generating one or more converted images and storing the original image and the one or more converted images as processing target images; and (c) detecting a fixed size that can be included in each of the processing target images. With respect to the area, by changing the position of the detection area in the processing target image, scanning means for scanning each of the processing target images, and (d) at each position of the detection area changed by the scanning means, Face detection means for extracting an image portion corresponding to a detection area from the processing target image and detecting a face area for the image portion, wherein the face detection means includes (d-1) the image portion. A high-luminance detection means for detecting a high-luminance part; and (d-2) a removal means for removing the high-luminance part from the image portion when the high-luminance part is detected by the high-luminance detection means. Have

  According to a third aspect of the present invention, in the face detection apparatus according to the first or second aspect of the present invention, the high-intensity detection unit includes a calculation unit that calculates an average luminance value for the image portion, and the calculation unit. Means for detecting the high-luminance region based on the calculated average luminance value.

  According to a fourth aspect of the present invention, in the face detection device according to any one of the first to third aspects, the detection area includes glasses when a face region is detected in the detection area. A predetermined area is defined, and the high-intensity detecting means has means for detecting the high-intensity part of the image portion corresponding to the predetermined area.

  Further, the invention of claim 5 is a face detection method for detecting a face region from an image, and relates to an acquisition step of acquiring an image, and a detection area of a certain size included in the image acquired in the acquisition step. By changing the position of the detection area in the image, a scan step for scanning the image, and an image portion corresponding to the detection area at each position of the detection area changed in the scan step from the image A face detection step of extracting and detecting a face region for the image portion, wherein the face detection step includes a high-intensity detection step for detecting a high-luminance part for the image portion, and the high-intensity detection step. And a removal step of removing the high-luminance portion from the image portion when the luminance portion is detected.

  The invention of claim 6 is a face detection method for detecting a face region from an image, wherein an acquisition step of acquiring an original image and one or more conversions obtained by converting the resolution of the original image acquired in the acquisition step A resolution conversion step of generating an image and storing the original image and the one or more converted images as processing target images; and a position of the detection area with respect to a detection area of a certain size that can be included in each of the processing target images. By changing in the processing target image, a scanning step of scanning each of the processing target images, and an image portion corresponding to the detection area at each position of the detection area changed in the scanning step, the processing target image And a face detection step for detecting a face area for the image portion, the face detection step for the image portion And high-intensity detecting step of detecting, when the high-luminance portion in the high-luminance detecting step is detected, and a removal step of removing the high luminance region from the image portion.

  According to the first to sixth aspects of the present invention, an image portion corresponding to the detection area is extracted from the image at each position of the detection area to be changed in the image, and the face region is detected for the image portion. When a high luminance part is detected from the part, the high luminance part is removed. As a result, it is possible to easily and accurately detect the face region even for an image in which reflected light from the glasses is reflected.

  In particular, in the inventions of claims 2 and 6, since one or more converted images obtained by converting the resolution of the original image are generated, face detection can be performed without depending on the size of the face in the original image.

  In the invention of claim 3, since the high-luminance part is detected based on the average luminance value calculated for the image portion, the face area can be detected appropriately according to the luminance characteristic of the image.

  Further, in the invention of claim 4, since a high-luminance part is detected for a part corresponding to a predetermined area where glasses can exist when a face region is detected in the detection area, a high-luminance part is detected. Can be detected efficiently.

<First Embodiment>
<Configuration of image processing apparatus>
FIG. 1 is a functional block diagram showing the configuration of the image processing apparatus 1A according to the first embodiment of the present invention.

  The image processing apparatus 1A is configured as a personal computer, for example, and includes an operation unit 11 that has a keyboard and a mouse and receives an input operation from a user, and a control unit 10A that is connected to the operation unit 11 so as to be able to transmit. . Further, the image processing apparatus 1A includes an image input unit 12, a face detection unit 13, and a storage unit 14 connected to the control unit 10A, and functions as a face detection device that detects a face area from an image.

  The image input unit 12 is a part that inputs a digital image taken by, for example, a digital camera, and functions as an acquisition unit that acquires an image. For example, a recorded image is read from a memory card and image input is performed.

  The face detection unit 13 is a part that performs face detection, which will be described in detail later, on an image of a person photographed.

  The storage unit 14 is a part that stores an image or the like input to the image input unit 12.

  The control unit 10A includes a CPU and a memory, and is a part that performs overall control of the operation of the image processing apparatus 1A.

<About face detection>
In the image processing apparatus 1A, the face area is detected by the face detection unit 13 for the image input to the image input unit 12, and this process will be described below.

  FIG. 2 is a diagram for explaining face detection.

  Processing for changing the position of a window (detection area) WD of a certain size included in the image GI along the arrow direction DR with respect to the input image GI, that is, a window having a size corresponding to a part of the image GI The entire image GI is scanned by raster scanning the WD. Note that the luminance value (pixel value) of the image GI has 256 gradations. In the above scanning operation, an image portion corresponding to the window WD is extracted from the image GI at each position of the window WD in the image GI, and a face area is sequentially detected for the extracted image portion. Thereby, when a human face FC enters the window WD, it can be detected as a face area.

  However, when a person in the image wears spectacles and light such as external light is reflected on the spectacles, as shown in FIG. 3A, the eye portion is a high luminance part (for example, white) HB. As a result, the accuracy of face detection is significantly reduced.

  Therefore, the face detection unit 13 converts the luminance value (pixel value) of each pixel constituting the high-luminance part, that is, low luminance when the high-luminance part exists in the image in the window WD. Process.

  Specifically, first, it is determined whether the luminance value P (n) of each pixel in the window WD is larger than the threshold value Th1, that is, whether the following equation (1) is satisfied.

  P (n)> Th1 (1)

  Here, since the reflected light component of the glasses has a very high luminance value, for example, 250 is set as the threshold Th1.

  With the above determination operation, it is possible to appropriately detect a high luminance part.

  For the pixel satisfying the above equation (1), the luminance value P (n) is changed to α as in the following equation (2). As this α, a low luminance value, for example, 0 is set.

  P (n) = α (2)

  When the high luminance part HB corresponding to the eye part shown in FIG. 3A is detected by the above processing, it is converted into a low luminance part LB composed of low luminance (for example, black) pixels. It will be. As a result, the light component (high luminance part) in the face detection target image is appropriately removed, so that the eye part is in an appropriate state, and face detection can be performed with high accuracy.

<Operation of Image Processing Apparatus 1A>
FIG. 4 is a flowchart showing the basic operation of the image processing apparatus 1A. This operation is executed by the control unit 10A.

  In step S1, an image to be subjected to face detection processing is input to the image input unit 12. Here, for example, the image GI shown in FIG. 2 is input.

  In step S2, the image GI input in step S1 is scanned with the window WD shown in FIG.

  In step S3, it is determined whether there is a remaining image portion that has not been scanned in step S2. For example, as shown in FIG. 2, when scanning is started from the upper left of the image GI, it is determined whether the window WD has reached the lower right of the image GI. If there is a remaining image portion, the process proceeds to step S4, and if not, the process proceeds to step S6.

  In step S4, the light component is removed from the image portion in the window WD (described later).

  In step S5, it is determined whether a face exists in the image in the window WD from which the light component has been removed in step S4. In this face determination, for example, the face area of a person is detected by the method described in Patent Document 1 described above.

  In step S6, face selection processing is performed. For example, when a plurality of faces are detected at substantially the same position in the image GI in step S5, processing for selecting one as the face of the same person is performed.

  FIG. 5 is a flowchart showing the light component removal operation corresponding to step S4.

  In step S11, a pixel to be processed is selected in the image portion in the window. As a result, the pixels are sequentially processed.

  In step S12, it is determined whether there is a remaining pixel that has not been selected in step S11 in the image portion in the window. That is, it is determined whether all the pixels in the window have been selected. Here, if there are remaining pixels, the process proceeds to step S13, and if not, the process proceeds to step S5.

  In step S13, it is determined whether the luminance value P (n) of the pixel selected in step S11 is larger than the threshold value Th1 as in the above equation (1). By this operation, as described above, high-luminance pixels in the window can be detected. Here, when the luminance value P (n) of the pixel is larger than the threshold value Th1, the process proceeds to step S14, and otherwise, the process returns to step S11.

  In step S14, α is substituted for the luminance value P (n) of the pixel selected in step S11. That is, the high luminance pixel is changed to the low luminance α (for example, 0) as in the above equation (2).

  The above-described operation of the image processing apparatus 1A detects and removes a high-luminance part such as reflected light from spectacles, so that face detection can be performed easily and accurately. That is, since the pixel value corresponding to the reflected light component is corrected using the property that the luminance value of the reflected light component reflected on the glasses is very high, it is possible to prevent the face detection accuracy from being lowered.

  In the image processing apparatus 1A, the face position may be specified by performing face selection processing as follows in step S6 shown in the flowchart of FIG.

  First, the center coordinate position of a window determined as a face is determined as the representative coordinate position of the window. When a face is detected at a plurality of window positions, an average coordinate position obtained by averaging these representative coordinate positions is calculated, and the average coordinate position is determined as a face position.

  As described above, the face position can be appropriately determined also by the face selection process.

Second Embodiment
An image processing apparatus 1B according to the second embodiment of the present invention has a configuration similar to that of the image processing apparatus 1A according to the first embodiment shown in FIG. 1, but a face detection unit is different.

  That is, the face detection unit 13B of the image processing apparatus 1B performs the light component removal operation described below.

<About light component removal>
FIG. 6 is a flowchart showing the light component removal operation. In the face detection process of the image processing apparatus 1B, the same operation as that in the flowchart in FIG. 4 is performed. However, in the operation in step S4, the light component removal operation shown in the flowchart in FIG. 6 is performed.

  In step S21, an average luminance value Ave for all pixels in the window is calculated.

  In steps S22 to S23, operations similar to those in steps S11 to S12 shown in the flowchart of FIG. 5 are performed.

  In step S24, it is determined whether the luminance value P (n) of the pixel selected in step S22 satisfies the condition of the following expression (3).

  P (n)> Ave + Th2 (3)

  In the above equation (3), Ave is the average luminance value calculated in step S21, and Th2 is a predetermined threshold value.

  With the above determination operation, a high-luminance portion can be appropriately detected based on the average luminance value Ave. In step S24, if the above equation (3) is satisfied, the process proceeds to step S25. If not satisfied, the process returns to step S22.

  In step S25, α (eg, 0) is substituted into the luminance value P (n) of the pixel selected in step S22, as in step S14 of FIG.

  Since the image processing apparatus 1B detects and removes a high-luminance part such as reflected light from spectacles, face detection can be performed easily and accurately. That is, since the reflected light reflected on the glasses is determined using the average luminance value in the window, the reflected light component can be corrected in accordance with the luminance characteristics in the window, and the face detection accuracy can be prevented from being lowered.

<Third Embodiment>
An image processing apparatus 1C according to the third embodiment of the present invention has a configuration similar to that of the image processing apparatus 1A according to the first embodiment shown in FIG. 1, but a face detection unit is different.

  That is, the face detection unit 13C of the image processing apparatus 1C performs the light removal processing operation described below.

<About light component removal>
FIG. 7 is a diagram for explaining light component removal of the image processing apparatus 1C.

  In the face detection, a process for determining whether or not a face exists in the window WD is performed. If a face is detected in the window WD, the position of the part, that is, the position of the eyes, nose, mouth, etc. Can be roughly predicted.

  Therefore, in the image processing apparatus 1C, instead of detecting high-luminance pixels that satisfy the condition of the above expression (1) for all the pixels in the window WD, a face is detected in the window WD as shown in FIG. Two areas E1 and E2 in which glasses can exist are defined, and high luminance pixels are detected for each pixel constituting the image portion corresponding to these areas E1 and E2.

  As described above, since the high luminance part is detected only in the areas E1 and E2 in the vicinity of the center of the window WD, the reflected light component in the window WD can be detected efficiently, and rapid processing becomes possible.

  In the image processing apparatus 1C having such a configuration, operations similar to those in the flowcharts of FIGS. 4 and 5 are performed. However, in step S11 of FIG. 5, each pixel in the areas E1 and E2 shown in FIG. 7 is selected. The Rukoto.

  The operation of the image processing apparatus 1C described above provides the same effects as those of the first embodiment. Furthermore, since the reflected light component is detected only in the region where the glasses can exist in the window, this high-luminance part can be detected efficiently.

<Fourth embodiment>
An image processing apparatus 1D according to the fourth embodiment of the present invention has a configuration similar to that of the image processing apparatus 1A according to the first embodiment shown in FIG. 1, but the face detection unit and the control unit are different.

  That is, the face detection unit 13D and the control unit 10B of the image processing apparatus 1D perform the operation of the image processing apparatus 1D described below.

<Operation of Image Processing Device 1D>
FIG. 8 is a flowchart showing the basic operation of the image processing apparatus 1D.

  In step S31, the same operation as step S1 shown in the flowchart of FIG. 4 is performed.

  In step S32, a multi-resolution image is created. Specifically, an image obtained by sequentially reducing the image input in step S31 at a predetermined magnification is created. That is, one or more converted images obtained by converting the resolution of the original image acquired in step S31, for example, three reduced images GIa to GIc having different resolutions from the original image GI as shown in FIG. 9 are generated. The reduced images (converted images) GIa to GIc and the original image GI are stored in the storage unit 14 as processing target images for which face detection is performed.

  In step S33, a window is scanned for the processing target image. Specifically, the window WN is sequentially scanned with respect to the four images GI and GIa to GIc having different resolutions shown in FIG. Since the window WN has a fixed size that can be included in each of the images GI and GIa to GIc, the area ratio of the window WN to each of the images GI and GIa to GIc is different as shown in FIG.

  In step S34, it is determined whether there is a remaining image portion that has not been scanned in step S33. That is, the window scan is performed for each of the processing target images GI and GIa to GIc, and it is determined whether or not the scan has been completed for all the image portions in the last image. If there is a remaining image portion, the process proceeds to step S35, and if not, the process proceeds to step S37.

  In steps S35 to S37, operations similar to those in steps S4 to S6 shown in the flowchart of FIG. 4 are performed.

  The operation of the image processing apparatus 1D described above provides the same effects as in the first embodiment. Furthermore, since each image having different resolution (multi-resolution image) is scanned with a window of a fixed size, face detection can be performed without depending on the size of the face in the original image.

  In the face selection process in step S37 of FIG. 8, in addition to averaging the representative coordinate positions of a plurality of windows in which faces are detected as described above, each scale in the multi-resolution image is also averaged. It is preferable. This will be specifically described below.

  For example, consider a case where the size of the original image is 320 × 240 pixels and the window size is 20 × 20. When a face is detected in an image having a resolution of 160 × 120, the size of the face in the original image is 40 × 40. Further, when a face is detected in an image having a resolution of 100 × 75, the size of the face in the original image is 64 × 64. In this way, when faces are detected at two different resolutions and the positions are almost the same, the two sizes are averaged, and the size of 52 × 52 is determined as the final face size in the original image.

  As described above, by averaging each scale in the multi-resolution image, it is possible to detect a face having an appropriate size.

1 is a functional block diagram illustrating a configuration of an image processing apparatus 1A according to a first embodiment of the present invention. It is a figure for demonstrating face detection. It is a figure for demonstrating the removal of a light component. It is a flowchart which shows the basic operation | movement of 1 A of image processing apparatuses. It is a flowchart which shows the operation | movement of light component removal. It is a flowchart which shows the operation | movement of the optical component removal in the image processing apparatus 1B which concerns on 2nd Embodiment of this invention. It is a figure for demonstrating the optical component removal of 1 C of image processing apparatuses which concern on 3rd Embodiment of this invention. It is a flowchart which shows the fundamental operation | movement of image processing apparatus 1D which concerns on 4th Embodiment of this invention. It is a figure for demonstrating a multi-resolution image.

Explanation of symbols

1A to 1D Image processing apparatuses 10A and 10B Control unit 12 Image input units 13A to 13D Face detection units E1 and E2 Area FC face GI image in which glasses can exist (original image)
GIa to GIc Reduced image HB High luminance part LB Low luminance part WD, WN window

Claims (6)

A face detection device for detecting a face area from an image,
(a) acquisition means for acquiring an image;
(b) with respect to a detection area of a certain size included in the image acquired by the acquisition means, the scanning means for scanning the image by changing the position of the detection area in the image;
(c) At each position of the detection area changed by the scanning unit, a face detection unit that extracts an image part corresponding to the detection area from the image and detects a face region for the image part;
With
The face detection means includes
(c-1) high luminance detection means for detecting a high luminance portion for the image portion;
(c-2) When the high-intensity part is detected by the high-intensity detection unit, a removing unit that removes the high-intensity part from the image portion;
A face detection apparatus comprising: A face detection device for detecting a face area from an image,
(a) acquisition means for acquiring an original image;
(b) resolution conversion means for generating one or more converted images obtained by converting the resolution of the original image acquired by the acquisition means, and storing the original image and the one or more converted images as processing target images;
(c) With respect to a detection area of a certain size that can be included in each of the processing target images, a scanning unit that scans each of the processing target images by changing the position of the detection area in the processing target image;
(d) At each position of the detection area that is changed by the scanning unit, an image part corresponding to the detection area is extracted from the processing target image, and a face detection unit that detects a face region for the image part;
With
The face detection means includes
(d-1) high luminance detection means for detecting a high luminance portion for the image portion;
(d-2) When the high-intensity part is detected by the high-intensity detection unit, a removing unit that removes the high-intensity part from the image portion;
A face detection apparatus comprising: In the face detection device according to claim 1 or 2,
The high brightness detection means includes
Calculating means for calculating an average luminance value for the image portion;
Means for detecting the high luminance part based on the average luminance value calculated by the calculating means;
A face detection apparatus comprising: The face detection device according to any one of claims 1 to 3,
The detection area defines a predetermined area in which glasses can exist when a face area is detected in the detection area,
The high brightness detection means includes
Means for detecting the high-intensity portion of the image portion corresponding to the predetermined area;
A face detection apparatus comprising: A face detection method for detecting a face area from an image,
An acquisition process for acquiring images;
With respect to a detection area of a certain size included in the image acquired in the acquisition step, a scan step of scanning the image by changing the position of the detection area in the image,
A face detection step of extracting an image portion corresponding to the detection area from the image at each position of the detection area changed in the scanning step, and detecting a face region for the image portion;
With
The face detection step includes
A high-intensity detection step of detecting a high-intensity part for the image portion;
When the high-luminance portion is detected in the high-luminance detection step, a removal step of removing the high-luminance portion from the image portion;
A face detection method characterized by comprising: A face detection method for detecting a face area from an image,
An acquisition process for acquiring the original image;
A resolution conversion step of generating one or more converted images obtained by converting the resolution of the original image acquired in the acquisition step, and storing the original image and the one or more converted images as a processing target image;
A scan step of scanning each of the processing target images by changing the position of the detection area in the processing target image with respect to a detection area of a certain size that can be included in each of the processing target images,
A face detection step of extracting an image portion corresponding to the detection area from the processing target image at each position of the detection area changed in the scanning step, and detecting a face region for the image portion;
With
The face detection step includes
A high-intensity detection step of detecting a high-intensity part for the image portion;
When the high-luminance portion is detected in the high-luminance detection step, a removal step of removing the high-luminance portion from the image portion;
A face detection method characterized by comprising:

Images