JP2005301722A - Face region detector - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a face region detector by which excessively picking up a redundant region is inhibited to enhance face detection accuracy since most of the non-face region included in face candidate regions detected by the existing face candidate region detection technique is excluded to allow face and non-face determination to the remaining face candidate region only. <P>SOLUTION: The face region is detected by using the face region detector configuring the following: a picture input section 101 reading a picture; a face candidate region detection section 102 calculating a face feature to detect the face candidate regions; a photographing conditions information input section 103 reading photographing conditions information added to the picture; a face candidate region selection section 104 selecting a region highly likelihood as the face region out of the face candidate regions by using scene type information and photographing distance information; a face region decision section 105 determining to see whether or not each face candidate region includes a face; and a face region information storage section 106 recording face region information. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a method for detecting a face area from an image taken by a digital camera or the like.

  In recent years, with the rapid spread of digital cameras, mobile phones with camera functions, and the like, there are increasing opportunities for users to easily handle a large number of images. Users will organize the images they have taken later, but it takes a long time to visually organize the images one by one, so it is expected to develop a method that can automate the organization and classification. .

  When a user classifies an image, it is possible to classify it by a keyword describing the shooting date / time, shooting location, event name at the time of shooting, etc., but as a classification method that is most desired to be automated, it is a shooting target. There is a method of classifying by subject.

  If the subject is a person, the burden on the user can be further reduced if it is possible to automatically identify who the person is. For this purpose, it is necessary to first detect the face area in the image stably and then perform the process of recognizing the person, but the current technology has not established a face detection method at the initial stage. In many cases, the obtained face candidate area includes a background or the like.

  SUMMARY OF THE INVENTION An object of the present invention is to provide a face area detection device that detects a face area with higher accuracy, excluding false detections included in a face candidate area.

  FIG. 3 shows a technique using template matching, which is one of conventional face detection techniques (see, for example, Patent Document 1). In FIG. 3, 301 is an image input unit for reading an image, 302 is a feature amount extraction unit for extracting information for facilitating capturing facial features from the image, and 303 is a template input unit for inputting a face template created in advance. , 304 is a similarity calculation unit that calculates the similarity between the feature amount extracted while scanning the image and the template, and 305 is a face that detects an area satisfying a certain standard in the image and defines it as a face area It is an area extraction unit.

  However, the above-described conventional method cannot detect a face area included in a snapshot taken by a digital camera or the like without excess or deficiency. This is because the snapshot is a restriction like a person image taken for security (person is facing front, only one person is in the image, face position / size is always the same) This is because there are various face area patterns, and it is difficult for the template or other feature amount to imitate the facial features sufficiently.

As described above, it is difficult to accurately detect a face region only from image features.
JP 2001-222719 A

  The problem to be solved is that it is difficult to detect the face area included in the original image using only the image features.

  The most important feature of the present invention is to use shooting condition information recorded at the time of shooting an image in order to improve the accuracy of the face detection method using the image feature.

  According to the present invention, most of the non-face regions included in the face candidate regions detected by the existing face candidate region detection method can be excluded, and face non-face determination is performed only on the remaining candidate regions. Therefore, there is an advantage that it is possible to prevent the excessive area from being picked up and improve the face detection accuracy.

  Hereinafter, embodiments of the present invention will be described with reference to FIGS.

(Embodiment 1)
FIG. 1 is a configuration diagram of a face area detection apparatus according to Embodiment 1 of the present invention. In FIG. 1, 101 is an image input unit that reads an image, 102 is a face candidate region detection unit that calculates a feature amount from the image and detects a face candidate region, and 103 is a shooting that reads a shooting condition added to the image at the time of image shooting. A condition information input unit 104 is a face candidate region selection unit 105 that selects a region that meets the conditions input by the imaging condition information input unit 103 from the face candidate regions detected by the face candidate region detection unit 102. A face area determination unit 106 that determines whether the face candidate area selected by the face candidate area selection unit 104 is a face and determines a final face area, and 106 is determined by the face area determination unit 105 A face area information storage unit that records face area information on a recording medium.

  First, the image input unit 101 reads an image shot by a shooting device such as a digital camera. A face candidate area is detected by the face candidate area detection unit 102 from the read image. Further, the shooting condition information input unit 103 reads the shooting condition information recorded together with the image at the time of image shooting. Next, in the face candidate area selection unit 104, among the primary face candidate areas 151 detected by the face candidate area detection unit 102, secondary face candidate areas that match the conditions read by the imaging condition information input unit 103 152 is selected. Then, the face area determination unit 105 determines in detail whether or not the selected secondary face candidate area 152 is a face, and the final face area is determined. Finally, the determined face area information is recorded on the recording medium by the face area information storage unit 106. Here, the shooting condition information refers to scene type information that records the situation at the time of image shooting, shooting distance information that records the distance from the shooting device to the subject, and the like.

  An example of the face candidate area detection method applied by the face candidate area detection unit 102 will be described with reference to the flowchart shown in FIG. First, an edge feature is detected from the image (S701). Next, Fourier transformation is performed on the image in which the edge feature is detected (S702), and the image is transformed into a frequency feature. Then, a product sum with the frequency feature of the template image created in advance is calculated (S703), and inverse Fourier transform is performed to convert the frequency feature into two-dimensional coordinates (S704). Finally, an area where the sum of products calculated in S703 shows a certain threshold value or more is detected as a face candidate area (S705). The advantage of using the frequency feature is that the processing amount can be greatly reduced because the processing for obtaining the correlation can be simplified while scanning the entire area of the image as in normal template matching.

  An example of the face non-face determination method applied by the face area determination unit 105 will be described with reference to the flowchart shown in FIG. Based on the information of the secondary face candidate area output by the face candidate area selection unit 104, an image of the corresponding area is cut out and input as a face candidate area image (S401), and most of the face candidate areas are skin color areas. It is determined whether there is an eye area (S402) or whether the face candidate area has an eye area (S403), and a face candidate area that satisfies the condition is determined as a final face area.

  Details of the face candidate region selection unit 104 will be described below. FIG. 2 is a configuration diagram showing details of the face candidate region selection unit 104, and FIG. 5 is a flowchart of the configuration. In FIG. 2, 201 is an information extraction unit that extracts scene type information and shooting distance information from shooting condition information, 202 is a face image selection unit that excludes images that do not have a face area based on the scene type information, and 203 An imaging distance matching area selecting unit that excludes face candidate areas having a size that does not match the imaging distance information.

  Here, the scene type information is information recorded in accordance with the surrounding environment and the state of the subject at the time of image shooting, such as landscape mode and macro shooting mode. The shooting distance information is, for example, information on the focal length, the shooting magnification, the CCD size, the number of pixels, or the distance from the shooting device to the subject itself. When the distance from the photographing apparatus to the subject is recorded, the distance to the subject is directly recorded, or a rough distance range to the subject is simply recorded.

When the scene type information is set to landscape mode or macro shooting mode in the image selection unit 202 (S502, S503), it is assumed that no person is captured in the image, and all candidate regions in the image are excluded. (S504).
Next, the shooting distance matching area selection unit 203 excludes face candidate areas having a size that does not match the shooting distance information. As shown in FIG. 6, there are face candidate area sizes corresponding to the types of template sizes applied by the face candidate detection unit 102, and face candidate areas of various sizes are detected. As shown in FIG. 8, the correspondence table between the shooting distance and the face candidate area size at a certain resolution / magnification can be obtained in advance. It is assumed that the actual measurement value of the face candidate area size at each distance is always the same. FIG. 9 shows a model in which a CCD light receiving element as a light receiving surface forms an image. In FIG. 9, a is the focal length, b is the distance from the camera lens to the subject, c is one side of the actual measurement value of the face candidate region size, and x is one side of the face candidate region size on the CCD. Here, since a is added to the image as distance information, if b and c are defined as shown in FIG. 8, x can be calculated by the following mathematical formula.

If x is converted into a pixel unit based on the number of pixels per CCD unit, the face candidate area size on the image can be determined (S505). The face candidate area is limited by comparing the distance to the actual subject with this correspondence table. However, there are few common photographing apparatuses such as a digital camera having an accurate distance measuring function. However, when trying to focus on the subject by autofocus, the distance can be roughly estimated. The distance information to the subject obtained at this time is stored, for example, as information on whether the distance is a short distance or a long distance with respect to a certain reference distance. The reference distance varies depending on the photographing apparatus, but is a known value. Now, when the reference distance is set to 3 m and the distance information is set to a short distance, it is assumed that the subject is not farther than 3 m (S506), and a small area size corresponding to a distance farther than 3 m in the correspondence table. Are excluded from the candidates (S507). On the other hand, when the long distance is set, the face candidate area is narrowed down by removing the large area size corresponding to the distance closer than 3 m from the candidates (S508). In the case of an imaging apparatus capable of accurate distance measurement, only the region sizes that fall within the range near the corresponding distance are detected as candidate regions from the correspondence table, and other region sizes are excluded from the candidates.

  As described above, according to the present embodiment, most of the non-face area included in the face candidate area can be removed, so that the accuracy of the face area detection device can be improved.

  The face area detection apparatus according to the present invention is an apparatus configuration that not only improves face detection accuracy by excluding non-face areas from face candidate areas, but also performs face non-face determination processing after limiting the number of candidate areas. This also contributes to shortening the face non-face determination processing time.

Configuration diagram of face area detection apparatus according to Embodiment 1 of the present invention Configuration diagram of face candidate region selection unit in Embodiment 1 of the present invention Configuration of conventional face detection device Flow chart of face non-face determination method in face area detection apparatus in Embodiment 1 of the present invention Flow diagram of face candidate area selection unit in face area detection device in embodiment 1 of the present invention Explanatory drawing which shows the method of a suitable candidate area size range determination in the flowchart of the face candidate area | region selection part shown in FIG. Flow diagram of face candidate area detection method in face area detection apparatus in Embodiment 1 of the present invention The figure which shows the example of the face candidate area size and shooting distance in a certain resolution and magnification Explanatory diagram showing the relationship between the image on the CCD and the measured value

Explanation of symbols

DESCRIPTION OF SYMBOLS 101 Image input part 102 Face candidate area detection part 103 Shooting condition information input part 104 Face candidate area selection part 105 Face area determination part 106 Face area information storage part 151 Primary face candidate area 152 Secondary face candidate area 201 Information extraction part 202 Face Image selection unit 203 Shooting distance matching region selection unit 301 Image input unit 302 Feature amount extraction unit 303 Template input unit 304 Similarity calculation unit 305 Face region extraction unit

Claims (2)

An image input unit that reads an image, a face candidate region detection unit that detects a face candidate region by calculating an image feature, and shooting condition information that reads shooting condition information added to the image or set corresponding to the image A face candidate region selecting unit for selecting a region having a high likelihood as a face region from the face candidate regions using the shooting condition information read by the shooting condition information input unit; and each face candidate region A face area detection apparatus comprising: a face area determination unit that determines whether or not a face includes a face; and a face area information storage unit that records face area information. In the face candidate area selection unit, an information extraction unit that extracts scene type information that records a situation at the time of image shooting from shooting condition information and shooting distance information that records a distance from a shooting device to a subject, and scene type information A face image selection unit that excludes an image having no face area originally and a face candidate area selection unit that excludes a face candidate area of an inappropriate size based on shooting distance information The face area detection apparatus according to claim 1.

Images