JP2000067237A - Person identifying device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To properly detect the positions of eyes in spite of how much the eyes are opened, to properly detect the eyes and mouth while reducing the quantity of calculation and to improve identification accuracy. SOLUTION: This person identifying device is provided with an image input part 100 for inputting an image, face area detecting part 200 for detecting the area of a face image from the input image, a feature point detecting part 300 for detecting the feature points of a face based on the detected face image, a normalization part 400 for deforming the image so as to come to a preset position by moving the positions of the detected feature points and a recognition part 500 for identifying and recognizing a person while using the normalized image and the data-base images of respective registered persons registered beforehand. The feature point detecting part 300 has an eye detecting part 310 for detecting the feature points of the eyes while using the template of the eyes composed of a secondary curve showing the boundary on he upside of the eyes and a circle showing the pupils and a mouth detecting part 340 for finding the boundary area of upper and lower lips from the secondary differentiated image in the preset direction and defining the center of gravity as the feature point of the mouth.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a person identifying apparatus for identifying a person using a face image.

[0002]

2. Description of the Related Art In order to make a man-machine interface between a machine and a person more natural, it is necessary to specify a person from a face image. In addition, in order to perform identification with high accuracy, it is important to accurately normalize the face image.

Conventionally, in this type of person identification device,
Person identification is performed through the following processing steps. 1) Extract a face region from an image. 2) Extract feature points such as eyes and mouth. 3) Normalize the face image using the extracted feature points and affine transformation. 4) Identification is performed using pattern matching with a face image of a dictionary prepared in advance or using a neural network.

Here, for the detection of eyes and mouth, there has been proposed a method of extracting a feature point by using a template pattern prepared in advance and comparing it with a face image. For example, FIG. 5 shows an example of an eye template pattern, which includes a vertical distance a from the center of the eye to the upper boundary arc, a horizontal distance b from the center of the eye, and a center of the eye. , The vertical distance c to the lower boundary arc, the pupil radius r, the inclination of the eye θ, and the distance xc and yc from the center of the eye to the center of the eye. As another method, a method has been proposed in which a separation filter for extracting an area of a specific feature such as a circle or a rectangle is combined with template pattern matching.

[0005]

However, the above-mentioned prior art has the following problems. (1) The template pattern used in the conventional example has a large number of parameters, and the amount of calculation increases when the template pattern is made to correspond to the shape change and size of the target pattern. (2) In the conventional example, the method using the template pattern can be applied only to a relatively simple shape such as a circle or a rectangle, and cannot cope with a case where the eyes are closed. (3) The recognition accuracy is low because the detection error of the feature point is not taken into account.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problems of the prior art, and can correctly detect the position of the eyes regardless of the degree of opening of the eyes, reduce the amount of calculation, and correct the eyes and the mouth. It is an object of the present invention to provide a person identification device that detects and improves identification accuracy.

[0007]

In order to achieve this object, a person identifying apparatus according to the present invention comprises an image input means for inputting an image and an input image input via the image input means. A face area detecting means for detecting an area of the face image, a feature point detecting means for detecting a feature point of the face based on the face image detected by the face image area detecting means, and a feature point detected by the feature point detecting means. By moving the position of the feature point, a normalizing means for deforming the image so as to come to a preset position, a normalized image obtained from the normalizing means and a database image of each registrant registered in advance. And a recognizing means for recognizing and recognizing a person by using an eye template, wherein the feature point detecting means uses an eye template consisting of a quadratic curve indicating the upper boundary of the eye and a circle indicating the pupil. It has an eye detection unit for detecting, and a mouth detection unit that obtains a boundary region between the upper and lower lips from a second derivative image in a preset direction and uses the center of gravity as a feature point of the mouth. .

[0008]

FIG. 1 is a system block diagram showing a configuration of a person identification device according to an embodiment of the present invention.
As shown in FIG. 1, the person identification device according to the present embodiment includes an image input unit 100 for inputting an image, a face area detection unit 200 for detecting a face image area from the input image,
A feature point detection unit 300 that detects a feature point of the face based on the detected face image, and an image that is moved by moving the position of the feature point detected by the feature point detection unit 300 so as to come to a preset position. And a recognizing unit 500 for recognizing and recognizing a person using the normalized image obtained from the normalizing unit 400 and a database image of each registrant registered in advance.

Here, the face area detecting section 200 registers a standard face image in advance and registers a standard face image database section 210 with an input image and a standard face image registered in the standard face image database section 210. A correlation calculating unit 220 that calculates a correlation, and detects a region having a high correlation as a face region.

The feature point detector 300 receives an eye template generator 320 for generating an eye template including a quadratic curve indicating the upper boundary of the eye and a circle indicating the pupil, and a face area detector 200. And a correlation calculation unit 330 for calculating a correlation between the eye region and the generated eye template in the face region to be detected. An eye detection unit 310 for detecting as a feature point, a boundary region detection unit 350 for obtaining upper and lower lip boundary regions from a second derivative image in a preset direction, and a centroid calculation unit 36 for calculating the center of gravity
0, and a mouth detection unit 340 that detects the center of gravity obtained by the center-of-gravity calculation unit 360 as a feature point of the mouth.

The recognition section 500 further includes a personal face image database section 510 in which face images of each person are registered in advance, and a correlation between the face image of each person and the normalized image input through the normalization section 400. And a correlation operation unit 520 that calculates

Next, with respect to the operation of each part of the person identification device having the above configuration, an explanatory diagram of an example of changes in the eye template and parameters shown in FIG. 2, a mouth detection region and boundary regions and features of the upper and lower lips shown in FIG. This will be described with reference to the explanatory diagram of the point detection result and the explanatory diagram of the movement state of the feature point shown in FIG.

First, the image input unit 100 inputs a face image from a video camera or the like. In the face area detection unit 200,
By the correlation operation unit 220, for the input face image,
The standard face images registered in the standard face image database unit 210 are sequentially moved over the entire screen to detect a region having the highest correlation.

In the feature point extracting section 300, the face area detecting section 2
00, the right eye, the left eye, and the mouth are set at predetermined positions with respect to the face image detected at 00, and each eye generated by the eye template generation unit 320 is set for all the right eye and the left eye. The correlation calculation unit 33 calculates the correlation with the template
Then, the area having the highest correlation is detected, and the center of the eye template of the area is set as a feature point.

As shown in FIG. 2A, the eye template is composed of a quadratic curve indicating the upper boundary of the eye and a circle indicating the pupil. The coefficient a of the quadratic term of ax ² + Bx + C (B,
C: a constant uniquely determined from the values of a, b, and c), a value b that is half the width of the eye, a y-coordinate value c of the outer corner of the eye, and a radius r of the pupil.

Therefore, in the eye detection, since a simplified template is used, the number of parameters is small and the amount of calculation is small. Further, as shown in FIGS. 2B to 2D, the position of the eyes can be detected even when the parameters are changed. That is,
FIG. 2B shows a state in which the eyes are open, FIG. 2C shows a state in which the eyes are thin, and FIG. 2D shows a state in which the eyes are closed. Even if different, the position of the eyes can be detected.

As for the mouth detection, the mouth detection unit 34
In FIG. 3A, first, as shown in FIG. 3A, a horizontal boundary between upper and lower lips is obtained by a boundary region detection unit 350 in a preset detection region of the mouth of the original image, and the vertical direction of the image is determined. To obtain a boundary area larger than a preset threshold value. next,
A region whose vertical width is smaller than a preset threshold and whose horizontal length is the longest is selected from the obtained boundary regions, and the center of gravity of the region is obtained by the center-of-gravity calculation unit 360, which is used as a feature of the mouth. Point. FIG. 3B shows the detection results of the boundary regions of the upper and lower lips and the feature points at this time.

Next, the normalizing section 400 moves the feature points obtained by the feature point detecting section 300 by dx and dy pixels in the horizontal direction and the vertical direction, respectively, as shown in FIG. Affine transformation is performed so that the position comes to a preset position, and a normalized image is created.

Then, the recognizing unit 500 calculates the correlation between the image of each person registered in the personal face image database unit 520 and the normalized image by the correlation calculating unit 510, and determines the person of the image with the highest correlation. And

[0020]

As described above, according to the present invention, the feature points of the eye are detected by the feature point detecting means using the eye template composed of the quadratic curve indicating the upper boundary of the eye and the circle indicating the pupil. An eye detection unit for detecting, and a mouth detection unit that obtains a boundary area between the upper and lower lips from a second derivative image in a preset direction and uses the center of gravity as a feature point of the mouth, has the following effects. To play. (1) Since a simplified template is used in eye detection, the number of parameters is small and the amount of calculation is small. (2) The position of the eyes can be accurately detected regardless of the degree of opening of the eyes. (3) In mouth detection, the algorithm is simple and the amount of calculation is small. (4) High robustness against detection errors of feature points.

[Brief description of the drawings]

FIG. 1 is a system block diagram showing a configuration of a person identification device according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram for explaining the operation of the eye detection unit in FIG. 1 and showing an example in which an eye template and parameters are changed.

FIG. 3 is an explanatory diagram for explaining the operation of the mouth detecting unit in FIG. 1 and showing detection results of a mouth detection region of an original image, a boundary region between upper and lower lips, and a feature point.

FIG. 4 is an explanatory diagram of a moving state of a feature point in the present invention.

FIG. 5 is an explanatory diagram showing an example of an eye template used in a conventional example.

[Explanation of symbols]

 REFERENCE SIGNS LIST 100 image input section (image input section) 200 face area detection section (face area detection section) 210 standard face image database section 220 correlation operation section 300 feature point detection section (feature point detection section) 310 eye detection section 320 eye template generation section 330 Correlation calculation unit 340 Mouth detection unit 350 Boundary region detection unit 360 Center of gravity calculation unit 400 Normalization unit (normalization unit) 500 Recognition unit (recognition unit) 510 Correlation calculation unit 520 Personal face image database unit

Claims (1)

[Claims] 1. An image input means for inputting an image, a face area detection means for detecting a face image area from an input image input via the image input means, and a detection by the face image area detection means A feature point detecting means for detecting a feature point of the face based on the detected face image; and moving the position of the feature point detected by the feature point detecting means to deform the image so as to come to a preset position. And a recognition unit for identifying and recognizing a person using a normalized image obtained from the normalization unit and a database image of each registrant registered in advance, the feature point detection unit An eye detection unit that detects an eye feature point using an eye template consisting of a quadratic curve indicating the upper boundary of the eye and a circle indicating the pupil; and an upper and lower lip from a second derivative image in a preset direction. of Seeking field region, and a mouth detector that the centroid feature points of the mouth, the person identification device having.