JPH11312243A - Facial region detector - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a facial region detector, capable of securely detecting from an image a face without erroneously recognizing a region excepting for the face and the face from each other. SOLUTION: A detector is provided with a mosaicking part 2 for mosaicking an inputted image, a mosaicking part 4 for mosaicking an image through a filter 3 to obtain the absolute value of high frequency components of the inputted image, a standard face image data base 5 for previously storing templates T1 and T2 of the low/high frequency components of a standard image, correlation arithmetic parts 6 and 7 for obtaining correlation between the templates T1 and T2 by successively scanning images from the parts 2 and 4 for each previously fixed area obtained by dividing the images into prescribed sizes, a correlated value adding part 8 for adding corresponding mutually correlated values outputted from the parts 6 and 7 by multiplying a weighting coefficient, a maximum correlation detecting part 9 for obtaining a largest correlated value and its area from the correlated values and a correlation determining part 10 for determining the largest correlated value as the region of the face, when it is larger than a previously fixed threshold.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to preprocessing of an apparatus for identifying an individual using a face image, and more particularly to a face area detecting apparatus for detecting a face image from an image.

[0002]

2. Description of the Related Art In order to construct a more natural interface between a computer and a human, research has been conducted on human identification and facial expression recognition by a computer. In order to identify humans and recognize facial expressions, it is necessary to detect a face image from an image.

FIG. 2 is a system configuration diagram showing a conventional face image detecting device. As shown in FIG. 2, the conventional face image detection device mosaicizes an image input via an image input unit 11 and outputs a mosaic image P0.
2, a standard face image database 13 in which a standard face image template T0 is stored in advance, and an area of a preset size of the mosaic image P0 are sequentially scanned and stored in the area and the standard face image database 13. A correlation calculation unit 14 for obtaining a correlation with the template T0 of the obtained face image, a maximum correlation region detection unit 15 for detecting a region with the highest correlation from the image, and a case where the highest correlation value is larger than a preset threshold value And a correlation determining unit 16 for determining a face area.

That is, in the conventional face image detecting apparatus, first, as shown in FIG. 3, an input image of an original image input through an image input unit 11 is converted into m × n pixels ( N1 <n <N2, M1 <m <M2,
(n, m are integers) mosaic image P0 having a reduced resolution by performing a mosaic process with the average value of the signals in the region as one pixel.
Generate Then, as shown in FIG. 4, a region of a mosaic image having a predetermined size of M3 × N3 pixels is sequentially scanned, and the correlation operation unit 14 stores the region and the previously stored region stored in the standard face image database 13. The template T of the standard face image as shown in FIG.
A correlation with 0 is obtained, and a region having the highest correlation is detected from the image by the maximum correlation region detection unit 15. Assuming that the higher the correlation is, the larger the correlation value is. If the highest correlation value is larger than a preset threshold, the correlation determination unit 16 outputs the area as a face area.

[0005]

However, in the above-described conventional face image detecting apparatus, since high frequency components are lost in the mosaic processing, there is a problem that an area other than the face is easily erroneously recognized as a face.

SUMMARY OF THE INVENTION The present invention has been made in order to solve the above-described problems of the conventional example. In order to detect a face from an image, an area other than the face is accurately detected without being erroneously recognized as a face. It is an object of the present invention to provide a face region detecting device capable of performing the following.

[0007]

In order to achieve the above object, a face area detecting apparatus according to the present invention comprises: a first mosaic means for mosaicizing an input image including at least a face image; Filter means for determining the absolute value of the high-frequency component, second mosaic means for mosaicizing the image via the filter means, and a standard face image database for storing templates of low-frequency and high-frequency components of the standard face image in advance. A template of a low-frequency component of a standard face image stored in the standard face image database by sequentially scanning the image mosaiced by the first mosaic unit into predetermined regions each having a predetermined size; The first mosaic means for calculating the correlation with the image and the image mosaiced by the second mosaic means are divided into a predetermined size. Second sequentially scanned to every defined Luo beforehand region correlating with the template of the high-frequency components of the standard face image stored in the standard face image database
Correlation calculation means, correlation value addition means for adding corresponding correlation values output from the first and second correlation calculation means by multiplying them by a predetermined weighting factor, and output from the correlation value addition means Maximum correlation detection means for obtaining the largest correlation value and its area from the correlation value, and correlation determination means for determining the area as a face when the maximum correlation value obtained by the maximum correlation detection means is larger than a predetermined threshold. It is a thing.

[0008]

FIG. 1 is a system configuration diagram showing a face image detecting apparatus according to an embodiment of the present invention. As shown in FIG. 1, the face image detection device according to the present embodiment has an input image Y including a luminance signal input via the image input unit 1.
A first mosaic unit 2 for mosaicizing 0, a filter unit 3 for obtaining the absolute value of the high-frequency component of the input image Y0, a second mosaic unit 4 for mosaicizing the image via the filter unit 3, A standard face image database 5 in which templates of low frequency components and high frequency components of the standard face image are stored in advance, and a predetermined area obtained by dividing the image mosaiced by the first mosaic unit 2 into a predetermined size. , A first correlation calculator 6 for calculating the correlation between the low frequency component of the standard face image stored in the standard face image database 5 and the template, and an image mosaiced by the second mosaicizer 4. Is sequentially scanned for each of predetermined regions obtained by dividing the standard face image into a predetermined size. A correlation calculation unit 7 for calculating the correlation value; a correlation value addition unit 8 for multiplying the corresponding correlation values output from the first and second correlation calculation units 6 and 7 by a predetermined weighting coefficient; A maximum correlation detector 9 for determining the largest correlation value and its area from the correlation values output from the value adder 8,
A correlation judging unit for judging as a face area when the maximum correlation value obtained by the maximum correlation detecting unit is larger than a predetermined threshold.

That is, in the face image detecting device shown in FIG. 1, an image Y0 composed of a luminance signal is output from the image input unit 1, and the mosaic unit 2 outputs m × M as shown in FIG.
A mosaic image P0, P1, P2,... is newly created in which the average value of the signals in the area of n pixels is newly set to the value of one pixel. Here, m and n are integers taking all values in the range of M1 <m <M2, N1 <n <N2, and mosaic images P0, P1, P2,... Thereby, it is possible to cope with a change in the size of the face.

[0010] In the correlation calculating section 6, the area mosaic images P0, P1, P1 of M3 × N3 pixels having a predetermined size are set.
P2,... Are sequentially scanned, and the area and the previously stored FIG.
As shown in the figure, the correlation between the low-frequency component of the standard face image having the size of M3 × N3 pixels and the template T1 is obtained. Here, for example, FIG. 4 corresponding to the area where the upper left coordinate of the area of the image P0 is (xn, yn) shown in FIG.
(B) is expressed as rn (xn, yn).

On the other hand, the filter section 3 takes the absolute value of the signal after passing the input image Y0 from the image input section 1 through, for example, a band-limiting filter having the following characteristic.
In the following equation, r is the distance from the origin, p is a parameter related to the frequency band, G (r) is a function of the Gaussian distribution, and the parameter p related to the pass band of the filter is the size m, n of the mosaic area. Function.

[0012]

(Equation 1)

Therefore, the number of images F0, F1, F2,...
-Is output. The mosaic unit 4 performs the same processing as the mosaic unit 2 on the input images F0, F1, F2,. At this time, the mosaic processing is performed with the size of the area corresponding to the characteristics of the filter unit 3, and the mosaic images Q0, Q
1, Q2,... Are output.

As shown in FIG. 4, the correlation operation unit 7 includes mosaic images Q0, Q1, Q2,.
A region of a predetermined size of M3 × N3 pixels is sequentially scanned in the image, and the region and the high frequency of the standard face image stored in the standard face image database 5 as shown in FIG. Component template T2
Find the correlation with Here, for example, the correlation value shown in FIG. 4B corresponding to the region where the upper left coordinates of the region of the image Qn shown in FIG. 4A is (xn, yn) is represented by sn (x
n, yn).

The correlation value adder 8 multiplies the correlation values (rn (xn, yn) and sn (xn, yn)) of the mosaic images Pn and Qn of the same resolution with respect to the area at the same position by multiplying them by a certain weighting coefficient. Fit together. Maximum correlation detector 9
Assumes that the higher the correlation, the greater the correlation value.
The largest correlation value and a region corresponding to the largest correlation value are obtained from the correlation value output from the correlation value adding unit 8. When the maximum correlation value obtained by the maximum correlation detection unit 9 is larger than a preset threshold value, the correlation determination unit 10 determines that the region is a face region, and when the maximum correlation value is smaller, the face region does not exist in the image.

Therefore, according to the present embodiment, the following effects can be obtained. (1) Since not only low frequency components but also high frequency components via the filter unit 3 are used, the probability of erroneously recognizing a face other than a face as a face is small. (2) Since an image composed of high frequency components is used in a mosaic form, it is hardly affected by individual differences of the face, so that the face is rarely erroneously recognized as a non-face.

[0017]

As described above, according to the present invention, as pre-processing for identifying an individual using a face image,
Regarding a device for detecting a face region from an image, the image is divided into low-frequency components and high-frequency components, each image is mosaiced, a correlation is obtained with a template of a standard face image, and these are multiplied by appropriate weights. By performing addition and detecting a face area using the addition, it is possible to perform detection with high robustness without erroneously recognizing an area other than the face as a face.

[Brief description of the drawings]

FIG. 1 is a system configuration diagram showing a face area detection device according to an embodiment of the present invention.

FIG. 2 is a system configuration diagram showing a face area detection device according to a conventional example.

FIG. 3 is an explanatory diagram of a mosaic processing.

FIG. 4 is an explanatory diagram of scanning of a region for which a correlation is to be obtained and a corresponding correlation value.

FIG. 5 is an explanatory diagram showing an example of a face image template stored in standard face image databases 5 and 13;

[Explanation of symbols]

 Reference Signs List 1 image input unit 2 mosaic unit (first mosaic unit) 3 filter unit (filter unit) 4 mosaic unit (second mosaic unit) 5 standard face image database 6 correlation operation unit (first correlation operation unit) 7 correlation operation unit (second correlation operation unit) 8 correlation value addition unit (correlation value addition unit) 9 maximum correlation detection unit (maximum correlation detection unit) 10 correlation judgment unit (correlation judgment unit)

Claims (1)

[Claims] A first mosaic unit configured to mosaic an input image including at least a face image; a filter unit configured to determine an absolute value of a high-frequency component of the input image; and an image mosaiced through the filter unit A second mosaic unit, a standard face image database in which templates of low-frequency components and high-frequency components of the standard face image are stored in advance, and a pre-divided image obtained by dividing the image mosaiced by the first mosaic unit into a predetermined size. A first correlation calculating means for sequentially scanning every predetermined area to obtain a correlation with a template of a low-frequency component of the standard face image stored in the standard face image database; and a mosaic by the second mosaic means The scanned image is sequentially scanned for each predetermined area divided into a predetermined size and stored in the standard face image database. Second correlation calculating means for calculating a correlation between the stored high-frequency component template of the standard face image and a corresponding correlation value output from the first and second correlation calculating means by a predetermined weighting factor Correlation value adding means for calculating the largest correlation value and its area from the correlation value output from the correlation value adding means, and a maximum correlation value obtained by the maximum correlation detection means. And a correlation judging means for judging as a face area when the threshold value is larger than the threshold value.