JPH07302337A - Device for retrieving image of similar face - Google Patents

Abstract

PURPOSE:To attain the formation of a similar face data base suitable for similar face retrieval matched with human sensitivity and its retrieval, and to display a face image to be easily observed. CONSTITUTION:A face area to be compared for retrieval is constituted of a face part area such as an area including hair, an area not including hair, eyes and their peripheral part, a nose and its peripheral part, and a mouth and its peripheral part or the combination of respective parts and the face area is formed like mosaic by changing block size in accordance with the face parts and registered in a data base 300. At the time of retrieval, the mosaic of the face image obtained from an input face image by a means similar to a registering means is compared with the mosaic registered in the data base 300 and persons are selected and outputted in the order of similarity to the mosaic of the input face.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a similar face image retrieving device which cuts out a face from an input face image and stores the face portion in a database or retrieves a face most similar to the input face from this database. Is.

[0002]

2. Description of the Related Art In order to find a face in a screen or to express a face, the conventional technique generally focuses on the shape of a face image, searches the input image for a candidate having a matching shape feature, and shapes the face. I used to express with features. However, it is very difficult to extract the correct shape from the face image, and it was forced to impose various restrictions such as strict lighting conditions and uniform background.

Therefore, for example, Japanese Patent Application No. 3-344489
In "Image recognition device" of the issue, a method of finding a face by sparse and dense search using the multiple resolution of an image or expressing the face by a mosaic is proposed. In this image recognition device, in particular, the constraint condition is relaxed, the target image is accurately found in the screen, and the object image to be recognized is recognized in order to recognize the object based on this result. Roughly mosaic and scan the unknown image by using this as a feature of the target image at the time of search to find the approximate position and size. Next, by using the data obtained by finely mosaicing the object image to be recognized, the vicinity of the roughly obtained position is scanned to find the accurate position and size. Furthermore, a fine mosaic of the recognition target object image and the mosaic of the unknown image are matched. By the above means, the conventional problems could be basically improved, but in terms of application,
Specifically, in order to apply this to similar face retrieval, studies were left on the appropriate area of the face and the size of the mosaic.

[0004]

In the technique proposed in the above image recognition apparatus, it is possible to identify Lincoln from a mosaic image of Lincoln, for example, by using the grayscale or color information of coarse resolution for the facial feature instead of the shape. As described above, it is based on the fact that the characteristics of the target can be expressed even in an image with a significantly reduced resolution.

However, in actual application to the search for similar faces, it is necessary to find a face area or expression method that searches for faces that match or are close to human subjectivity. The present invention is
It is assumed that the position of the face in the screen has been found by the technical means of the above-mentioned image recognition device already proposed, and similarly, the grayscale or color information of coarse resolution is used instead of the shape for expressing the facial features. The purpose is to make it possible to create a database and search for it in a form suitable for more effective similar face search. In other words, it is an object to realize an effective search with an expression close to human subjectivity by extracting and mosaicing face areas suitable for searching for similar faces.

[0006]

SUMMARY OF THE INVENTION The present invention is an apparatus for retrieving a similar face image by cutting out a face portion from the screen,
When the position and size of the face are given, the face area to be compared includes a hair-containing area, a hair-free area, an eye and its peripheral area, a nose and its peripheral area, or a mouth and its peripheral area. A first means for extracting at least one of a part area and a combination thereof, and a second means for mosaicizing this face area based on a block size predetermined according to a face part and registering it in a database; A person most similar to the input face mosaic is compared by comparing the mosaic of the face image obtained from the input face image by the same means as the first and second means with the mosaic in the database registered by the second means. Is selected, or a third means for selecting and outputting persons in a similar order is a first feature.

The second means has a fourth means for mosaicing the face area to be compared and registering it in the database, and cutting out the head area to register the in-area image in the face image database. A third means is to select the person most similar to the mosaic of the input face, or to select the persons in order of similarity, and then display the head image of the selected person as the fourth image.
The second feature is to have a fifth means for taking out from the face image database registered by the means and outputting it.

[0008]

[Function] By defining the face region and appropriately mosaicing it as described above, it becomes possible to make a database of input face images in a form suitable for similar face search. It is possible to search efficiently and with high accuracy.

[0009]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of the similar face image search device of the present invention. In the figure, 100 is an input unit, 200 is a processing unit, 300 is a database, and 400 is an output unit.

The input section 100 is composed of a TV camera 101, a scanner 102, or the like. The processing unit 200 includes an input image buffer 201, a positioning unit 202, a positioning dictionary 203, a central region cutout unit 204, a database registration unit 205, a similar face mosaic unit 206 to be searched,
It comprises an input face mosaic section 207 and a search section 208. The database 300 has a face image file 301 and a mosaic face file 302.

A TV camera 101, a scanner 102 or the like is used from the input unit 100 to input a face image of a person to be searched as a similar face and transfer it to the input image buffer 201. Therefore, the content of the input image buffer 201 is a set of pixels represented by shades or colors.

The image in the input image buffer 201 is sent to the positioning unit 202, and the positioning unit 202 uses, for example, the multi-resolution which is shown in Japanese Patent Application No. 3-344489 "Image Recognition Device" of the prior art. By searching,
The center position and size of the face are obtained by referring to the positioning dictionary 203.

The center area cutout unit 204 first cuts out a head image area from the center position and size of the face thus obtained by the positioning unit 202. The database registration unit 205 registers, in the database 300, a head image in which unnecessary portions have been removed from the input image by cutting out the head image area, as a face image file 301.

FIG. 2 shows an example in which the head image area is cut out. In the input image shown in FIG. 2A, the + mark indicates the center position of the face. The center area cutout unit 204 cuts out a head image area as shown by a broken-line rectangular frame in FIG. 2A from the center position and size of the face obtained by the positioning unit 202, and the database registration unit 205 The head image is registered in the database 300 as a face image file 301. FIG. 2B shows an example of a face image registered as the face image file 301.

Next, the center area cutout unit 204 cuts out a face area required as a similar face according to the purpose from the center position and size of the face. The mosaic unit 206 changes the block size in this area according to the purpose and mosaics it, and the mosaiced data is stored in the database 300.
It is registered inside as a mosaic face file 302.

FIG. 3 shows an example of areas and block sizes for a mosaic face. Reference numeral 500 in the figure represents a face area to be mosaiced. FIG. 3A shows an example in which the size of the face area 500 is medium. FIG. 3B shows the face area 5.
This is an example when the size of 00 is increased. Figure 3
(C) is an example when the size of the face area 500 is made small. 3D shows an example in which a square is used as the shape of the face area 500, FIG. 3E is a rectangle as the shape of the face area 500, and FIG. Shown respectively.

Various block sizes for mosaic in these face areas 500 can be selected. Figure 3
FIG. 3G shows an example in which a large block size is adopted in the square face area 500, and FIG. 3H shows an inverse convex face area 500.
3 (g) is set to a medium size smaller than that in FIG. 3 (g). In FIG. 3 (i), the block size of the square face area 500 other than near the eyes is set to medium, and the block size is further increased only near the eyes. Each example shows a reduced size.

Which of these is adopted depends on what the human uses as the degree of similarity of the face. For example, FIG. 3 (a) is generally used for the region, FIG. 3 (f) is used if the contour of the face is emphasized, and the average block size is FIG. 3 (g). Is adopted, and if the eye area is emphasized, FIG. 3 (i) may be adopted. Also, the shape and block size of the face area such as the nose and its peripheral area, the mouth and its peripheral area, or a combination thereof are adaptively selected according to the purpose.

The above procedure is repeated for the number of persons to be registered, and the database registration of face images and mosaic faces is completed. Next, the face image of the person to be searched for the similar face is input in the same procedure, and the center region cutout unit 204 uses the center position and the size of this face to identify the face region required as the similar face. Cut out accordingly. Further, the mosaic unit 207 changes the block size in this area according to the purpose by the same procedure as the mosaic unit 206, and mosaics the area. That is, the mosaic unit 207 is the mosaic unit 2 described above.
06 may be used. Here, the face area to be mosaiced and the block size are the same as when the mosaic face file was registered.

This mosaic face data is sent to the search unit 208. The search unit 208 calculates the similarity between this data and each data read from the mosaic face file 302 in the database 300, and determines the mosaic face most similar to the input face mosaic data to the mosaic face file 30.
The face images corresponding to the mosaic face file 302 are extracted from the face image file 301 in the database 300 and sent to the output unit 400. Output part 4
00 sequentially displays face images as search results.

By the way, the similarity between mosaic faces is generally defined as follows. Input face mosaic data,

[0022]

[Equation 1]

Then, the mosaic data in the database is

[0024]

[Equation 2]

It is assumed that Here, N is the number of mosaics that represent one face, and each face is represented by a vector having these mosaics as elements. The similarity S between two vectors is given by the following equation from the formula in vector operation.

[0026]

[Equation 3]

This gives the cosine between two vectors and is the same as the correlation value. The closer the two mosaics are, the smaller the angle between the two vectors becomes, and S becomes a value close to 1.

Therefore, this is repeated for other mosaic face files 302 in the database 300,
S is extracted in the order closest to 1. For example, in the above embodiment, the face images to be searched are mosaiced for the various regions and block sizes to be mosaiced as described with reference to FIG. 3, and a plurality of types of mosaiced data are stored in the database 300. Mosaic face file 3
If it is registered as No. 02 and the kind of mosaiced data to be used at the time of search can be selected from a menu or the like, a desired similar face image can be easily selected particularly according to the purpose. become able to.
For example, the data that is mosaiced for the entire face area including hair and the data that is mosaiced for the non-hair area are separately stored in the database 300 for each search target.
Register it in and make it selectable from the menu. In this way, when it is necessary to search for a similar face for the entire face including hair, the search is performed by specifying that the data that has been mosaic-registered for the region including hair is used, and that the similar hairstyle is required. If not, it is possible to perform a search that specifies to use the data about the area that does not include hair.

[0029]

As described above, according to the present invention,
It is possible to search for similar faces that match human sensibilities by matching the sense of similar faces of humans with the selection criteria for the face area and mosaic block size. In addition, a face image (head image) in which the size and position of the face are aligned can be obtained from various face images having a face at an arbitrary position on the screen,
It is possible to obtain a face image that is easy for humans to see.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a similar face image search device of the present invention.

FIG. 2 is a diagram showing an example in which a head image area is cut out.

FIG. 3 is a diagram showing an example of a region and a block size for a mosaic face.

[Explanation of symbols]

 100 Input Unit 101 TV Camera 102 Scanner 200 Processing Unit 201 Input Image Buffer 202 Positioning Unit 203 Positioning Dictionary 204 Central Region Clipping Unit 205 Database Registration Unit 206 Mosaicing Unit 207 Mosaicing Unit 208 Searching Unit 300 Database 301 Face Image File 302 Mosaic Face file 400 Output section 500 Face area to be mosaiced

Claims (2)

[Claims] 1. A device for extracting a face part from a screen and searching for a similar face image, when a face position and size are given, a face region to be compared does not include a hair-containing region and a hair-containing region. A first means for extracting a face part region such as a region, an eye and its peripheral part, a nose and its peripheral part, a mouth and its peripheral part, or a combination thereof; Second means for mosaicing based on one or a plurality of predetermined block sizes and registering in a database, and a mosaic of face images obtained from input face images by means similar to the first and second means Comparing with the mosaic in the database registered by the second means, selecting the person most similar to the mosaic of the input face, or selecting and outputting the persons in the similar order.
And a similar face image search device. 2. The similar face image search device according to claim 1, wherein the second means mosaics the face area to be compared and registers it in the database, and cuts out the head area to extract the in-area image from the face image database. The fourth means for registering the head image of the selected person is selected after selecting the person most similar to the mosaic of the input face or after selecting the persons in a similar order. 4. A similar face image search device having a fifth means for taking out from a face image database registered by means of No. 4 and outputting it.