JPS61208185A - Recognizing device for feature part of face - Google Patents

Abstract

PURPOSE:To extract feature points with a small quantity of arithmetic by comparing a feature parameter value of each area of a binary image with the standard parameter value of a specific area and discriminating the area, and calculating presence positions of feature parts by using the relative position relation between the position information of the obtained area and a previously calculated feature part. CONSTITUTION:A face image inputted from an image input part 101 is stored in an image storage part 102 and converted by a binary coding part 103 into a binary signal. This binary image is labeled, area by area, by a specific face area detecting part 4 to calculate feature parameters (area, peripheral length, center of gravity, etc.) of each label (area), thereby detecting the specific area (e.g. eye, etc.) of the face on the basis of the parameters. Then, a feature part presence position calculating part 105 calculates the range of the presence position of each feature part by using a proportional expression which represents the relative position relation (which allows the two-dimensional calculation of the range of the presence of eyebrows by statistics when the positions and interval of, for example, eyes are known) among feature parts.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a feature point recognition portion of a physiognomy recognition system used for identifying human faces.

Recognition of facial feature points for traditional engineers is an old and new topic. So far, research on human face identification has been conducted, such as a man-machine system for classifying facial photographs and an examination of facial % signature parameters, but none of these have gone beyond the realm of research.

Two studies have since been published. One is a method that detects boundaries from a silhouette image of a profile and identifies facial features based on the line shape of the boundaries (The AutomaticRec
ogration of Human Faces f
rom ProfileSilhouetteffiJ
G, T, KAUFMAN efc...I
EEETrons, on SMC, vol, S
MC-6, A2, FEB.

1976, P P 113 121) oThe other method is to filter the image data of the front of the face to detect edges, detect the features of the edge shape, and extract individual feature points (
mouth, nose, eyes, etc.) (Sakai et al.,
“Computer-based facial photo analysis,” Journal of the Institute of Electronics and Communication Engineers.

April 1973, PP, 226-233).

Problems to be Solved by the Invention One of the main objectives of the present invention is to obtain the positions of each feature point such as the eyes, nose, mouth, etc. and the region in which they exist from facial image data. The obtained results become the main parameters for physiognomic identification. In the above technology, when a face silhouette image is used, area information (existence range, shape, etc.) of each feature point is
is missing when inputting the image, and does not meet the above main purpose. Furthermore, in the case of a feature point detection method using a filtered image, a filtering operator with a large mask size is required in order to obtain region information, and a huge amount of calculation is required to apply this to the entire image size. Furthermore, in the case of a filtered image, it becomes sensitive to shadows appearing in the direction of the face. As a result, in the case of a face image in which the person is wearing glasses or the like, it becomes difficult to identify the eyes due to the shadows and frames of the glasses.

Means for Solving the Problems In order to solve the problems, the present invention sets an appropriate threshold value to generate a binary image including areas such as eyes, hair, eyebrows, etc., and in the resulting area, Select specific feature points that stably appear depending on the conditions at the time of image input, and define the area of these feature points as
Characteristic parameters of each area (area, perimeter, center of gravity coordinates, etc.)
It is used for identification.

Function: The method for recognizing facial features of the present invention stores image data of a human face inputted from an image input device in an image memory;
Binarize the stored image data to generate a binary image that includes a specific area of the face, calculate the characteristic parameter values of each area in the generated binary image, and compare them with the standard parameter values of the specific area of the face. By identifying the area, and using the obtained positional information of the area and the relative positional relationship of each characteristic part of the face calculated in advance, the position of each characteristic part of the face is calculated. The position of each characteristic part and the shape of the region are detected.

Embodiment FIG. 1 is a diagram showing an outline of an example of an apparatus for carrying out the present invention. In the figure, 101 is an image input section. The facial image input through the image input unit 101 is stored in the image storage unit 102. The stored face image data is stored in the binarization unit 103.
The image is then binarized and this binary image is temporarily stored. Label each region of the binary image, calculate characteristic parameters (area, perimeter, center of gravity, etc.) for each label (region), and detect specific (eye) regions of the face based on these. The facial specific area detection unit 104 does this. Next, the relative positional relationship between the positional information of the specific area obtained and the facial features calculated in advance (for example, if the position and spacing of the eyes are known, the range where the eyebrows should be statistically, the range where the mouth should be The characteristic part existence position calculation unit calculates the range of the existence position of each characteristic part of the face using a proportional expression expressing the following.

In this embodiment, a specific region of a face is used as an eye, and a recognition method will be specifically described below.

FIG. 2 is an example of face image data binarized using a threshold value θ. As shown in the figure, a concise binary image is obtained, and the eye shape remains realistic.

FIG. 3 shows the relationship between the threshold value θ using a histogram (frequency distribution of concentration). The concentration level at the first valley from the left in the figure is the threshold value θ. When an image is input with a bright single color background under constant lighting conditions, a histogram with clear changes as shown in FIG. 3 is obtained. Of these, eyes, black hair, etc. correspond to the level range marked ■ in the figure. ■The level range includes
This corresponds to partial areas of the face other than those mentioned above. The level range (2) corresponds to the background area. This tendency hardly changes even for hundreds of images when the above conditions are maintained. Furthermore, the value of θ is also approximately constant. Using this property, the level range of ■ is binarized and extracted as shown in FIG.

Initially, the data in FIG. 2 is a collection of pixels and has no meaning as a multi-region. Labeling involves combining these adjacent pixels and extracting a region. Note that labeling is described in detail in the reference document (P, H. Winston et al., LIsP, Baifukan, PP, 133-136).

FIG. 4 shows the result of numbering each region after labeling, and FIG. 6 shows the result of calculating the characteristic parameters for each label. Eye region pairs are detected by first applying a single eye condition to these feature parameters, and then applying a pair of eye conditions.

Next, the single eye condition and pair condition are shown.

Single eye condition ■ Sl < Area < 82 (Lower limit area) (Upper limit area) ■ Area > k1 × Perimeter long eye pair condition ■ 1y1-y21 <ΔY (Limit of vertical difference) ■ Δx1<1x1-x21<Δx2 ( Lower limit of left-right difference) (Upper limit of left-right difference) The eye area obtained as above is shown in FIG.

Using the line connecting the centroids of the detected eye areas as the base axis (L in Figure 7), the range of existence of each characteristic part is defined as an eyebrow mask, an eye mask, and a nose mask as shown in Figure 7.

I'll call it Lomask. The edges of each mask are parallel or perpendicular to the axis, and the mask itself is rectangular. In this case, the distance between the edge of each mask and the axis is shown between the arrows in Figure 7, and all values are expressed as a constant times (the length of the axis) 0. This shows that the relative positional relationship (distance) of each characteristic part does not differ greatly depending on the individual.

Therefore, if the positions of the centers of gravity of both eyes are known, the eyebrow mask, eye mask, nose mask, lobster mask, etc. can be automatically calculated. Furthermore, using the above mask, it is possible to apply i-specific processing to each characteristic part. Using the above method, an eyebrow mask is created directly above the binary values in Fig. 2.

Figure 8 shows the generation of an eye mask, a nose mask, and a nose mask.

Effects of the Invention As described above, the present invention is able to reduce the amount of calculation by using a binary image for identification.

In addition, by selecting the binarization threshold, it is possible to obtain a realistic region shape, and when identifying regions,
The original shape and features can be used for the discriminant function. Furthermore, by using the relative positional relationship of each feature, it is possible to obtain the existence range of feature parts other than the detected feature area, and using the range This method allows processing suitable for each characteristic portion to be performed.

[Brief explanation of drawings]

Figure 1 is a block diagram showing an overview of an example of a device for carrying out the present invention, Figure 2 is a diagram showing an example of an image obtained by binarization, and Figure 3 is a histogram of the original image data in Figure 2. Figure 4 is a diagram showing the results of labeling the image in Figure 2 and numbering each area, Figure 5 is a diagram showing the relationship between
Figure 6 shows the results of calculating the feature parameters for each area in the figure. Figure 6 shows the results of identifying the eyes from the image in Figure 2. Figure 7 shows the relative positional relationship of each characteristic part based on both eyes. FIG. 8 is a diagram showing the calculated existence range (mask) of each characteristic portion on the image of FIG. 2. 101... Image input unit, 102... Image storage unit, 103... Binarization unit, 104...
. . . Eye position detection section, 106 . . . Characteristic part existence position calculation section. Name of agent: Patent attorney Toshio Nakao and 1 other person 2nd
Figure 4 Figure 5 Rishin Figure 6 Figure 7 Figure 8

Claims (3)

[Claims] (1) means for storing image data of a human face input from an image input device in an image memory; means for binarizing the stored image data to generate a binary image including a specific region of the face; means for identifying the specific area by calculating the feature parameter value of each area in the binary image and comparing it with standard parameter values prepared in advance indicating the characteristics of the specific part of the face; Recognition of facial feature parts, characterized by comprising means for calculating the location of each feature part of the face using region position information and a relative positional relationship of the feature parts of the face calculated in advance. Device. (2) The facial feature recognition device according to claim 1, wherein the specific region is an eye. (3) The facial feature recognition device according to claim 1, wherein the feature parameters are the area, center of gravity coordinates, and perimeter of each region in the binary image.