JP2006092191A - Apparatus for detecting face image, method for detecting face image, and program for detecting face image - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an apparatus, a method and a program for detecting face images that make it possible to accurately and easily specify the positions of a number of face images in captured images in a short period of time, the face images being combined into a "backdrop" template image, etc. <P>SOLUTION: To continue the detection of face images, a face detection area setting part 11 sets as a detection area A an area obtained when an area Al' where detected face images exist is excluded from a currently set face detection area Al. A score threshold setting part 13 sets a score threshold V at V=Vl. There are no particular restrictions on the score threshold Vl to be reset as long as it is smaller than the currently set score threshold V. Face detection is carried out again based upon the newly set area A and the score threshold V. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a face image detection apparatus, a face image detection method, and a face image detection program, and more particularly, to an apparatus, method, and program for detecting a face portion of a person from a captured image in which a large number of persons are recorded.

Conventionally, various techniques for easily synthesizing a face image, which is an image of a face portion of a person image, a background image, and a clothing image have been developed. For example, according to Patent Document 1, while specifying two points as a reference for synthesis in a background image, a clothing image, and the like, the hair region of the face image and the inside of the face outline are used as a region for synthesis, Specify two points. Two points serving as a reference for the synthesis of the face image are specified so that the midpoint of the line connecting the two points is on the tip of the jaw on the horizontal line passing through the tip of the jaw and the length is the width of the face. Then, the portrait image is generated by mapping the region used for the synthesis of the face image so that the two points designated for the face image and the two points designated for the background image overlap.
Japanese Patent Laid-Open No. 10-222649

  By the way, recently, the character character's face part is blanked out, and a so-called “skin” -like template is created by inserting a face image extracted from an image obtained by photographing a plurality of persons into the blank part. Images are being developed. When a captured image such as a group photo of a large number of people is used as the face image extraction source, it is necessary to specify the positions of all face images from the captured image. However, due to the limitations of face detection technology, it is difficult to accurately specify the positions of all face images in one face detection. For this reason, it is necessary to manually repeat the selection of the erroneously detected object from the detected face image and the correctly detected face image. This operation becomes very complicated as the number of face images increases, and the load on the operator is expected to increase. The present invention has been made in view of such problems, and can accurately and easily identify the positions of a large number of face images to be combined with a “crisp” template image in a short time from a captured image. An object of the present invention is to provide a face image detection apparatus and method.

  In order to solve the above-described problems, a face image detection apparatus according to the present invention detects a face portion of a person in a captured image, a captured image input unit that inputs a captured image in which a person is recorded, an operation unit that receives an operation input, and the like. An area setting unit for setting a face detection area, a detection unit for detecting a human face part in the set face detection area, and calculating a score that quantifies the accuracy of detection of the detected face part And a score threshold setting unit for setting a threshold of a score serving as a reference for determining whether or not an object other than the face portion has been erroneously detected, and comparing the set threshold with the calculated score. A determination unit that determines a detected face part that is a face part with a score exceeding, a display unit that displays the detected face part, and an erroneous detection that is an object other than the face part that is erroneously detected from the detected face part Specify the object using the operation unit And a storage unit for storing the position of the detected face part other than the specified erroneously detected object, and the region setting unit starts from the currently set face detection region when repeating the detection. An area excluding the area where the detected face part exists is set as a new face detection area, and the score threshold setting unit sets a value less than the currently set threshold as a new threshold when the detection is repeated.

  According to the present invention, an area obtained by removing the area where face detection has been completed from the currently set face detection area is reset as a new face detection area, and the score threshold is a value smaller than the currently set score threshold Then, face detection is performed for the face detection area. By reducing the score threshold and repeating face detection for the undetected area of the face part in the captured image, the area with the undetected face part can be narrowed down, and a large number of people are recorded in the captured image. Even in this case, it is possible to specify the face image efficiently and accurately in a short time.

  The face image detection apparatus may further include a face size threshold value setting unit that sets a threshold value of the size of the face portion detected by the detection unit.

  By doing so, the size of the undetected face portion can be narrowed down.

  In order to solve the above-described problems, a face image detection method according to the present invention includes a captured image input step for inputting a captured image in which a person is recorded, and face detection that is an area for detecting a human face portion in the captured image. An area setting step for setting an area; a detection step for detecting a face part of a person in the set face detection area; a calculation step for calculating a score that quantifies the accuracy of detection of the detected face part; and a face A score threshold setting step for setting a threshold of a score serving as a reference for determining whether or not an object other than the portion is erroneously detected is compared with the calculated threshold and the face portion having a score exceeding the threshold A determination step for determining a detected face part, a display step for displaying the detected face part, and a false detection object that is an object other than the face part erroneously detected from the detected face part A false detection designation step for designating, and a storage step for storing the position of the detected face part other than the designated false detection object, and the region setting step when the detection is repeated, the currently set face detection region The area excluding the area where the detected face part exists is set as a new face detection area, and the score threshold setting step sets a value less than the currently set threshold as the new threshold when the detection is repeated. .

  This face image detection method has the same effect as the above-described face image detection device.

  In order to solve the above-described problems, a face image detection program according to the present invention includes a captured image input step for inputting a captured image in which a person is recorded, and face detection that is an area for detecting a human face portion in the captured image. An area setting step for setting an area; a detection step for detecting a face part of a person in the set face detection area; a calculation step for calculating a score that quantifies the accuracy of detection of the detected face part; and a face A score threshold setting step for setting a threshold of a score serving as a reference for determining whether or not an object other than the portion is erroneously detected is compared with the calculated threshold and the face portion having a score exceeding the threshold A determination step for determining a detected face portion, a display step for displaying the detected face portion, and a false detection that is an object other than the face portion that is erroneously detected from the detected face portions. An error detection designation step for designating an object and a storage step for storing the position of a detected face part other than the designated false detection object are executed by the computer, and the region setting step is currently set when the detection is repeated. An area obtained by removing the area where the detected face part exists from the face detection area being set is set as a new face detection area, and the score threshold setting step newly sets a value less than the currently set threshold when the detection is repeated. Set to a proper threshold.

  This face image detection program has the same effects as the above-described face image detection apparatus. The face image detection program may be provided by being recorded on a CD-ROM, DVD, MO, or other computer-readable recording medium.

  According to the present invention, an area obtained by removing the area where face detection has been completed from the currently set face detection area is reset as a new face detection area, and the score threshold is a value smaller than the currently set score threshold Then, face detection is performed for the face detection area. By reducing the score threshold and repeating face detection for the undetected area of the face part in the captured image, the area with the undetected face part can be narrowed down, and a large number of people are recorded in the captured image. Even in this case, it is possible to specify the face image efficiently and accurately in a short time.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings.

[Schematic configuration]
FIG. 1 is a schematic functional configuration diagram of a face image detection apparatus 100 according to a preferred embodiment of the present invention. The face image detection apparatus 100 includes a captured image input unit 1, a display unit 9, an operation unit 10, a processing unit 20, and a storage unit 30. The captured image input unit 1 inputs captured images acquired by a digital still camera, a film scanner, a media driver, various wireless / wired communication devices, and the like. The processing unit 20 is configured by a CPU, a one-chip microcomputer, and the like, and performs various arithmetic processes. The storage unit 30 is configured by a semiconductor memory, and stores a program executed by the processing unit 20 and data necessary for various operations of the processing unit 20. The display unit 9 is a display device configured with a liquid crystal display or the like. The operation unit 10 includes a keyboard, a touch panel, and the like that receive user operation inputs.

  Face detection area setting unit 11, score calculation unit 12, score threshold setting unit 13, face size threshold setting unit 14, face detection unit 15, detection result determination unit 16, detection result display unit 17, erroneous detection instruction unit 18, and face The extraction unit 19 is a program stored in the storage unit 30 and is loaded into the processing unit 20 and executed.

  The face detection area setting unit 11 sets an area (face detection area) in which the face detection unit 15 performs face detection on the captured image. The face detection unit 15 detects an image (face image) representing the face portion of a person from the face detection area by a known face recognition technique such as pattern matching. When a plurality of persons are recorded in the face detection area, a plurality of face portions are individually detected. The face size threshold value setting unit 14 sets an upper limit threshold value (face size upper limit value) or a lower limit threshold value (face size lower limit value) of the face size (specified by the number of pixels) detected by the face detection unit 15. The face detection unit 15 detects a face portion having a size smaller than the face size upper limit value when the face size upper limit value is set. On the other hand, when the face size lower limit value is set, the face detection unit 15 detects a face portion having a size smaller than the face size lower limit value. When neither the face size upper limit value nor the face size lower limit value is set, the face detection unit 15 detects face portions of all sizes.

  The score calculation unit 12 calculates a score obtained by quantifying the accuracy of detection of the human face portion detected by the face detection unit 15. In order to quantify the accuracy of detection of the face portion, for example, the circularity of the detected face portion can be used. Since the person's face is almost elliptical, a higher score is calculated when the circularity of the extracted face portion is close to the circularity of the ellipse, and a lower score is calculated as it deviates from the circularity of the ellipse. The score threshold setting unit 13 sets a score threshold that serves as a reference for determining whether the detected face part is actually a person's face part or an object other than the person's face part. Hereinafter, this threshold value is referred to as a score threshold value. The method of setting the score threshold is arbitrary, and an arbitrary value input from the operation unit 10 can be used as the score threshold.

[Process flow]
Next, a flow of face detection processing executed by the face image detection apparatus 100 will be described based on the flowchart of FIG. In the following processing, an area obtained by removing the face part instructed to be erroneously detected in the loop (steps S4 to S9) that has been executed and the face part that has been detected from the face detection area is set as the next face detection area. In addition, a score threshold lower than the score threshold set in the previous loop is set, and the detection of the face portion in the newly set face detection area is continued.

  In S1, the captured image input unit 1 inputs a captured image. Further, the score calculation unit 12 sets the number of executions l (l = 1, 2,...) Of step S4 to l = 1. FIG. 3 is an example of a photographed image input to the photographed image input unit 1. In the captured image, a plurality of persons Fn (n = 1, 2,..., 6) and an object X other than the person's face are recorded. The face portion of the person Fn is represented by fn.

  In S2, the score threshold setting unit 13 sets an initial value of the score threshold V. The initial value is not particularly limited, and an arbitrary value input from the operation unit 10 may be set as the initial value. For example, when the score of the face part fn is predicted to be between 0 and 200, the score threshold V = 100 is set as the initial value.

  In S3, the face detection area setting unit 11 sets the entire area A1 of the captured image as the initial face detection area A. A part obtained by excluding predetermined vertical and horizontal margins (margins) from the entire captured image may be set as the face detection area A.

  In S <b> 4, the face detection unit 15 detects a face part existing in the face detection region set by the face detection region setting unit 11. The score calculation unit 12 calculates a score for each detected face part.

  In S5, the detection result determination unit 16 compares the score calculated for each face part with the score threshold value V set by the score threshold value setting unit 13, and detects the face part whose score exceeds the score threshold value V. It is determined as an image (hereinafter referred to as a detected face image). The detection result display unit 17 displays the detected face image on the display unit 9. FIG. 4 is a display example of a detected face image. In this figure, among the persons F1 to F6 in FIG. 3, the face images f1 to f3 of some persons F1 to F3 that the face detection unit 15 has detected, and the persons falsely detected as face images by the face detection unit 15. An object X other than the face portion is extracted from the captured image and displayed. FIG. 4 is merely an example. In short, in the present embodiment, it is assumed that the face detection unit 15 can detect only a part of a face image, or that an object that is not a face part is erroneously detected as a face image.

  In S <b> 6, an instruction is received from the operation unit 10 as to whether or not the detection of the face portion is finished. If an instruction to finish the detection of the face portion is input from the operation unit 10, this process is terminated. If an instruction to end the detection of the face portion is not input from the operation unit 10, the process proceeds to S7. Instead of receiving an instruction on whether or not to detect the face part from the operation unit 10, whether to end the process or shift to S7 may be determined according to the number of face parts that have been detected. That is, the number of persons photographed in advance in S1 is input from the operation unit 10, and in S6, it is determined whether or not the number of detected face images matches the input number of persons. If it does not match, the process may be terminated.

  In S <b> 7, the erroneous detection instructing unit 18 causes the operation unit 10 to specify an object other than the face portion of the person mistakenly detected as the face image. The designation method is arbitrary. For example, the pointer P is clicked on an object other than the face portion with a pointing device. In this way, it is possible to easily exclude an object that is erroneously detected from the detected face image.

  FIG. 4 shows a state where an object X other than the face portion is designated as an erroneously detected object. The face extraction unit 19 stores, in the storage unit 30, the position on the captured image of the detected face image that is not designated as an erroneously detected object. For example, in FIG. 4, the positions of the face portions f <b> 1 to f <b> 3 are stored in the storage unit 30. Note that the face extraction unit 19 extracts a face image existing at a position stored in the storage unit 30. The processing unit 20 may synthesize the extracted face image by arranging it at predetermined positions of various template images.

  In S8, the score calculation unit 12 increments l (adds 1 to the current l). The face detection area setting unit 11 sets an area obtained by removing the area Al ′ where the detected face image exists from the currently set face detection area Al as the detection area A. Taking the detection results of FIGS. 3 and 4 as an example, the region A1 ′ is a circumscribed rectangle of the face portion that is determined to have been detected and is not designated as an erroneously detected object, and is newly set as the detection region A. The area A2 is an area obtained by removing the area A1 ′ from the currently set area A1. The center of the new detection area A may be designated from the operation unit 10.

  The method of setting the detection area A newly set in S8 is not particularly limited to the above. For example, as shown in FIG. 5, when the face portions f1 and f2 of two persons who are determined to have been detected and are not designated as erroneously detected objects are arranged in the lateral direction X of the captured image, it is further omitted. Since it is presumed that another person is lined up f1 to f2 at equal intervals, the center line m passes through a position X0 separated from the face parts f1 to f2 by a distance a in the horizontal direction X, and the person's face. The band region Rx extending along the vertical direction Y having a width W such that the horizontal direction can be accommodated may be set as a new region A. Alternatively, as shown in FIG. 6, when the face portions f1 and f2 of two persons who are determined to have been detected and are not designated as false detection objects are aligned in the vertical direction Y of the captured image, f1 Since it is presumed that another person row is photographed in the row between f2, the band region Ry extending along the horizontal direction X between the two persons f1 to f2 is set as a new region A. Also good.

  In S9, the score threshold setting unit 13 sets the score threshold V to V = Vl. The score threshold value V1 to be reset is not particularly limited as long as it is a value smaller than the score threshold value V currently set. For example, the score threshold value Vl is the sum of the scores of the face portions that are determined to have been detected and not designated as false detection objects, and the score threshold value Vl The sum is divided by a number larger than the number N (such as N + 1). Then, the process returns to S4 again, and face detection is performed based on the newly set area A and the score threshold value V.

  In S9, the face size threshold setting unit 14 may reset the face size threshold. For example, an average face size of a face portion that is determined to have been detected and is not designated as an erroneously detected object may be calculated, and the size may be set as the face size upper limit value or the face size lower limit value. Alternatively, a value larger than the currently set face size upper limit value may be set as a new face size upper limit value, or a value smaller than the currently set face size lower limit value may be set as a new face size lower limit value. May be set. By doing so, the size of the undetected face portion can be narrowed down.

  A face image detection program that causes the processing unit 20 to execute steps S <b> 1 to S <b> 9 is stored in the storage unit 30. The face image detection program may be recorded on a CD-ROM, DVD, MO, or other computer-readable recording medium and provided to the processing unit 20.

  As described above, the face image detection apparatus 100 resets an area obtained by removing the area where face detection has been completed from the currently set face detection area as a new face detection area (S8), and sets the score threshold to the current value. After resetting to a value smaller than the set score threshold V, face detection is performed for the face detection area. By reducing the score threshold V and repeating face detection for the undetected area of the face part in the captured image, the area with the undetected face part can be narrowed down, and a large number of persons are recorded in the captured image. Even in such a case, it is possible to extract a face image efficiently and accurately in a short time.

Schematic functional block diagram of face image detection device Flow chart showing the flow of face detection processing Figure showing the appearance of the captured image The figure which shows the detection result of the face image The figure which shows an example of the face detection area extended in the vertical direction The figure which shows an example of the face detection area extended in a horizontal direction

Explanation of symbols

1: shot image input unit, 9: display unit, 10: operation unit, 11: face detection area setting unit, 12: score calculation unit, 13: score threshold setting unit, 15: face detection unit, 16: detection result determination unit , 17: detection result display unit, 18: erroneous detection instruction unit, 19: face extraction unit

Claims (4)

A captured image input unit for inputting a captured image in which a person is recorded;
An operation unit for receiving operation inputs;
An area setting unit for setting a face detection area that is an area for detecting a human face portion in the captured image;
A detection unit for detecting a face portion of a person in the set face detection region;
A calculation unit that calculates a score obtained by quantifying the accuracy of detection of the detected face part;
A score threshold value setting unit for setting a threshold value of a score serving as a reference for determining whether or not an object other than the face part is erroneously detected;
A determination unit that compares the set threshold value with the calculated score and determines a detected face part that is a face part having a score exceeding the threshold value;
A display unit for displaying the detected face part;
An erroneous detection designating unit that causes the operation unit to designate an erroneously detected object that is an object other than the face part erroneously detected from the detected face part;
A storage unit for storing the position of a detected face part other than the designated erroneously detected object;
With
When repeating the detection, the region setting unit sets a region obtained by removing the region where the detected face part exists from the currently set face detection region as a new face detection region,
When the detection is repeated, the score threshold setting unit sets a value less than the currently set threshold as a new threshold.   The face image detection apparatus according to claim 1, further comprising a face size threshold setting unit configured to set a threshold value of a size of the face portion detected by the detection unit. A captured image input step for inputting a captured image in which a person is recorded;
An area setting step for setting a face detection area which is an area for detecting a face portion of a person in the captured image;
A detection step of detecting a face portion of a person in the set face detection area;
A calculation step of calculating a score obtained by quantifying the accuracy of detection of the detected face part;
A score threshold setting step for setting a threshold of a score serving as a reference for determining whether or not an object other than the face portion is erroneously detected;
A determination step of comparing the set threshold value with the calculated score and determining a detected face portion that is a face portion having a score exceeding the threshold value;
A display step for displaying the detected face portion;
An erroneous detection designation step for designating an erroneously detected object that is an object other than the erroneously detected facial part from the detected facial part;
A storage step of storing a position of a detected face part other than the designated erroneously detected object;
Including
In the region setting step, when the detection is repeated, a region obtained by removing the region where the detected face portion exists from the currently set face detection region is set as a new face detection region,
The score threshold setting step is a face image detection method in which, when the detection is repeated, a value less than a currently set threshold is set as a new threshold. A captured image input step for inputting a captured image in which a person is recorded;
An area setting step for setting a face detection area which is an area for detecting a face portion of a person in the captured image;
A detection step of detecting a face portion of a person in the set face detection area;
A calculation step of calculating a score obtained by quantifying the accuracy of detection of the detected face part;
A score threshold setting step for setting a threshold of a score serving as a reference for determining whether or not an object other than the face portion is erroneously detected;
A determination step of comparing the set threshold value with the calculated score and determining a detected face portion that is a face portion having a score exceeding the threshold value;
A display step for displaying the detected face portion;
An erroneous detection designation step for designating an erroneously detected object that is an object other than the erroneously detected facial part from the detected facial part;
A storage step of storing a position of a detected face part other than the designated erroneously detected object;
To the computer,
In the region setting step, when the detection is repeated, a region obtained by removing the region where the detected face portion exists from the currently set face detection region is set as a new face detection region,
In the score threshold setting step, when the detection is repeated, a face image detection program that sets a value less than the currently set threshold as a new threshold.

Images