JP2849256B2 - Image recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image recognizing apparatus which searches for a target image from input natural images with high accuracy and specifies an object.

[0002]

2. Description of the Related Art A typical natural image is a gray-scale image or a color image. However, in order to detect and recognize a target image from the natural image, conventionally, attention has been paid to the shape of the image, and the shape information is characterized. In this method, candidates having the same characteristics are searched for and recognized from the input image. However, it is difficult to extract the correct shape from a natural image.Conventionally, in order to separate the target object from the background, it is necessary to pre-learn the background or limit the background to a uniform one. And various restrictions such as restricting the target to artificial rigid objects or the like in order to easily extract line segments.

[0003]

Even the detection of the presence or absence of a target image is accompanied by such difficulties, and it is almost impossible to grasp the exact position and size of a target object in a natural image.

That is, it is extremely difficult to accurately extract a target object from an arbitrary natural image, and therefore, there are many restrictions in practical use, and the usefulness is difficult. SUMMARY OF THE INVENTION It is an object of the present invention to alleviate the above-mentioned various constraints, to accurately determine a target image in a screen, and to recognize a target based on the result.

[0005]

According to the present invention, a feature of an object is not a shape but a shade of coarse resolution or color information. That is, it is based on the fact that the feature of the target can be expressed even in an image with a greatly reduced resolution so that Lincoln can be identified from the Lincoln mosaic image.

More specifically, an object image to be recognized is roughly converted into a mosaic, and using this as a feature of the target image at the time of searching, an unknown image is scanned to find an approximate position and size. Next, the object image to be recognized is finely tessellated, and by using this, the vicinity of the already obtained rough position is scanned to find the correct position and size. further,
The recognition target object image is finely mosaiced, and the target image is recognized by matching the mosaic of the region in which the position and the size of the already obtained unknown image have been obtained with the data obtained by finely mosaicing the previous recognition target object image.

[0007]

By mosaicizing an image in several stages in this manner, it is possible to easily search for an object from an arbitrary screen, and to easily recognize the object.

[0008]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing an embodiment for performing image recognition according to the present invention. FIG. 2 is a diagram showing a configuration of a coarse search unit shown in FIG.
3 shows the configuration of the detailed search unit shown in FIG. 1, and FIG. 4 shows the configuration of the recognition unit shown in FIG. FIG. 5 and FIG. 6 together show a single figure, which is a diagram for explaining a process of converting a search target image or an unknown image into a mosaic, and searching and recognizing a target object from the unknown image.

In FIG. 1, reference numeral 1 denotes a recognition target image input unit which captures a face image of an object to be recognized, for example, a typical person in the case of a person, using a TV camera or a scanner. Specifically, it is input to the coarse search target image buffer 101 in FIG. Therefore,
The content of the buffer 101 is a set of pixels represented by shading or color.

This recognition target image is represented by a matrix F = [f]. As shown in FIGS. 5 and 6, this is divided into M × N by a block of Wa pixels × Wa pixels. each block represents a color having in the block, for example, if gray block average value, the maximum frequency in the block if the color. Note that the average value in the block
Represented by the color with the highest frequency in the
Values are referred to as “representative values,”
"Average value" of multiple blocks for multiple blocks
And expressed separately.

The coarse search target image mosaic unit 102 converts the image in the buffer 101 into a mosaic and stores it in the coarse search target image dictionary 103. If necessary, an average value of a plurality of search targets, in this example, a plurality of face images, may be taken as representative mosaic data of a face, or a plurality of coarse search target images may be prepared.

On the other hand, the unknown image U is taken into the unknown image buffer 104 shown in FIG. 2 in the rough search unit 100 via the unknown image input unit 2 shown in FIG. Therefore, as shown in FIG.
In step 5, the image is divided into P × Q with a block size of Wa pixels × Wa pixels, a representative value is calculated for each block, and a mosaic image UC is calculated.
Get. The unknown image input unit 2 may be the same as the recognition target image input unit 1 described above.

Next, as shown in FIG. 5, the mosaic of the unknown image is scanned to search for a place that matches the coarse search target image mosaic. That is, it is based on the following principle. When the arbitrary M × N mosaic data among the P × Q mosaic data of the unknown image is defined as Ui (indicated by an area with oblique lines), Ui
The position of Ui when the distance Di between the target and the coarse search target mosaic data FC is the minimum value is defined as the coarse search result. There are various ways of setting the distance Di, but a typical example is the Euclidean distance.

The details are as follows. In FIG. 2, reference numeral 106 denotes a candidate selection unit which selects candidate data Ui (i = 1, 2,...) From the unknown image mosaic unit 105 that falls within the frame of M × N blocks. Generally, first, the upper left M × N block of the unknown image is selected as U1. The mosaic data U1 and the coarse search target mosaic data FC from the coarse search target image dictionary 103 are input to the distance calculation unit 107, and the above distance is calculated. This result is stored in the position / size / distance storage unit 108 together with the M × N position data in the unknown image mosaic data UC at this time, that is, the position data of U1 and the mosaic block size Wa.

Further, the candidate selection unit 106 selects the next candidate U2 from the unknown image mosaic unit 105, and the distance calculation unit 107 calculates the distance at this time, and the position / size together with the position and block size of the candidate U2.・ Distance accumulation unit 108
To store. Similarly, M × N candidates Ui are sequentially selected for the entire region of the unknown image mosaic, and the selected position, block size, and calculated distance are all the position, size,
The distance is stored in the distance storage unit 108.

By the way, in the mosaic of the unknown image, the same block size as the image to be searched, that is, Wa pixels × Wa pixels is used. If the size of the search target image and the size of the search target image in the unknown image are the same, the size may be kept as it is, but generally, the size of the search target image in the unknown image is also unknown. For this reason, the unknown image mosaic unit 10
5 is the number of pixels of the block size, for example, 1 / Wa.
It is changed to Wa ′ such as 5 to 10/5. Therefore, the above procedure is repeated for each block size Wa ', and the obtained distance, the position at that time, and the block size Wa' are stored in the position / size / distance storage unit 108.

When a plurality of search target images are stored in the dictionary, the above procedure is similarly repeated for each search target image, and the obtained distance, the position at that time, and the block size are determined.・ Size / distance accumulation unit 108
To store.

Thus, when the distances in all cases are stored in the position / size / distance storage unit 108, the minimum value of the minimum distance detection unit 109 and the minimum value of the distance data in the position / size / distance storage unit 108 is calculated. Detect what to take. This value is sent to the recognition target presence / absence determination unit 110. If this value is less than a certain threshold value, it is determined that the target image exists in the unknown image, and its position and block size are extracted from the position / size / distance storage unit 108. The result is output to the coarse search result output unit 111. The coarse search result output unit 111 determines the size of the search target image in the unknown image from the block size Wa by [M × W
a] × [N × Wa].

On the other hand, when the minimum distance value exceeds a certain threshold value, it is determined that there is no target image in the unknown image, and "no" is output to the coarse search result output unit 111. The coarse search result output unit 111 immediately notifies the detailed search unit 200 of these results.

When the approximate position and size of the recognition target are known by the coarse search, the mosaic block is reduced in size and the position is searched in detail. In the detailed search, the area of the target image may be narrowed down to a part as compared with the coarse search. However, here, the case where the same area is used will be described.

First, as shown in FIG.
The block is divided into M ′ × N ′ by a block of b pixels × Wb pixels (Wb <Wa), and each block is represented by a representative value in the block.

More specifically, the detailed search target image buffer 2
01 fetches recognition target image data from the recognition target image input unit 1 and converts it into a detailed search target image mosaic unit 20
Send to 2. The detailed search target image mosaic unit 202 converts the image into a mosaic (FD) and stores the image in the detailed search target image dictionary 203.

If necessary, an average value of a plurality of detailed search target images or a plurality of detailed search target images may be prepared, as in the case of the coarse search. On the other hand, when the coarse search result is notified to the coarse search result notifying unit 220, the unknown image buffer 204 knows this and sends the unknown image buffer 204 from the unknown image input unit 2 of FIG. Import the relevant part. Therefore, the unknown image mosaic unit 2
On the other hand, in FIG. 5, as shown in FIG. 6, a block size of Wb pixels × Wb pixels is divided into P ′ × Q ′, and a representative value is calculated for each block to obtain a mosaic UD.

Therefore, as in the case of the coarse search, as shown in FIG. 6, the mosaic of the relevant partial image of the unknown image is scanned to search for a place that matches the detailed search target mosaic FD. That is, when an arbitrary M ′ × N ′ mosaic among the P ′ × Q ′ mosaic UDs of the partial image is defined as Uj (indicated by an area where oblique lines exist), the distance Dj between Uj and the mosaic FD to be searched Let the position of Uj when is minimized be the detailed search result.

More specifically, in FIG.
06 is M ′ × from the unknown image mosaic unit 205.
A candidate Uj that falls within the N ′ frame is extracted. Distance calculator 20
7 calculates the distance Dj of Uj and the detailed search target mosaic FD from the detailed search target image dictionary 203, and stores it in the position / size / distance storage unit 208 together with the position data of Uj and the mosaic block size Wb. send.

Here, if necessary, the block size Wb
Is changed (Wb ′) or a plurality of detailed search target images are used in the same manner as in the coarse search. When the minimum value is detected by the minimum distance detection unit 209 in this way, the position data of Uj and the block size Wb at this time are sent to the detailed search result output unit 210. The detailed search result output unit 210 calculates the exact size of the search target image from the block size Wb, that is, [M ′ × Wb] × [N ′ × Wb], and sends it to the recognition unit 300 together with the position data.

When the precise position and size of the object to be recognized are found by the detailed search, the mosaic block is reduced in size to recognize what the object is. First, as shown in FIG. 6, the recognition target is divided into M × N by a block having a size of Wc pixels × Wc pixels (Wc <Wb). Each block of the mosaic is represented by a representative value in the block as in the above case. If the feature to be recognized is concentrated in a partial region, that portion may be divided into M "× N". That is, in the process of searching, a universal feature is required for the object, but in the process of recognition, a feature for distinguishing the individual is required, and a finer mosaic or partial area is effective.

Here, the case where this partial area is used will be described. More specifically, the recognition target image buffer 301 takes in data of a partial area of the recognition target image from the recognition target image input unit 1 and converts the data into a recognition target image mosaicizing unit 302.
Send to The recognition target image mosaic unit 302 converts the image into a mosaic (FDD) and stores the mosaic in the recognition target image dictionary 303.

Normally, many images to be recognized are mosaiced and stored in the image dictionary for recognition 303. On the other hand, when the detailed search result is notified from the detailed search unit 200 to the detailed search result notifying unit 320, the unknown image buffer 304 knows this and sends the information from the unknown image input unit 2 in FIG.
An image of the exact position and size of the recognition target image in the unknown image U is captured. Image cutout / mosaic unit 305
Further, for the partial region therein, as shown in FIG. 6, a block size of Wc pixels × Wc pixels is M ″ × N ″.
And mosaic data UDD is obtained by calculating a representative value for each block.

Therefore, the distance between the mosaic data UDD and the recognition target image mosaic data FDD is calculated. If the distance is equal to or smaller than a certain threshold value, it is determined that the unknown image is the recognition target image. Alternatively, if there are a plurality of recognition target images, F
The object having the minimum distance from the DD is determined as a recognition target. Here, the latter case will be described as an example.

More specifically, in FIG.
06 is the unknown image mosaic data UDD output from the image cutout / mosaicization unit 305 and the recognition target image dictionary 3
The distance of the recognition target image mosaic data FDD from the target image 03 is calculated, and the distance and the code of the recognition target image are sent to the distance storage unit 307. Similarly, the distance calculation unit 306 takes out the recognition target image mosaic data one after another from the recognition target image dictionary 303, calculates the distance to the unknown image mosaic data UDD, and stores this value and the code of the recognition target image in the distance. Send to unit 307.

Next, upon detecting the minimum value in the distance accumulation unit 307, the minimum distance detection unit 308 sends the code of the image to be recognized at this time to the recognition result output unit 309. This result is further sent to the output unit 3.

On the other hand, as a result of the rough search, the target image is "none".
In this case, this is notified to the output unit 3. In this way, the approximate position of the recognition target image F is first found from the unknown image U, and the accurate position and size are found based on the approximate position, and the target is further specified.

[0034]

According to the present invention, a recognition target image and an unknown image are mosaiced and searched and recognized, so that no special condition such as uniformity with respect to the background is given. It is possible to easily find a target image, obtain its position and size, and recognize it without causing an error due to noise. It is particularly effective for objects having similar shapes and colors, such as human faces and butterflies.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating the whole of the present invention.

FIG. 2 is a diagram showing the contents of a coarse search unit 100.

FIG. 3 is a diagram showing contents of a detailed search unit 200.

FIG. 4 is a diagram showing the contents of a recognition unit 300.

FIG. 5 is a diagram illustrating a process of converting a recognition target image and an unknown image into a mosaic, and searching for and recognizing a recognition target in the unknown image stepwise.

FIG. 6 is a diagram illustrating a process of composing one diagram together with FIG. 5, mosaicizing a recognition target image and an unknown image, and gradually searching for and recognizing a recognition target from among the unknown images. .

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 recognition target image input unit 2 unknown image input unit 3 output unit 100 coarse search unit 200 detailed search unit 300 recognition unit

Claims (8)

(57) [Claims] 1. A natural image to be recognized is first stored in a first
Large block size corresponding to the block size (width W
a, height Wa) , dividing the image into M × N, calculating a representative value of shades or colors of each block, and storing the representative value as dictionary data . 2 blocks
Small block size corresponding to Kusaizu (width Wb, high
A second means for dividing into M ′ × N ′ by Wb, where Wb <Wa) , calculating a representative value of shading or color of each block, and storing the calculated value as dictionary data; Image in third block
A smaller block size corresponding to the size (width W
c, a height Wc, where Wc <Wb) , dividing into M "× N", calculating a representative value of shading or a color of each block, and storing it as dictionary data; When given as an input, the input image is divided into P × Q with a large block size corresponding to the first block size, a representative value of each block is calculated, and an arbitrary M × N area is taken out of this. A fourth means for calculating and accumulating a distance between the representative value group obtained by the first means, and a fourth means for repeatedly applying the fourth means over the entire area of P × Q. When the minimum value is obtained and the minimum value is smaller than a predetermined threshold value, it is determined that there is an image corresponding to the recognition target image at the position where the minimum value is derived in the input image, while the minimum value is larger than the threshold value. When the input image corresponds to the recognition target image A fifth means for determining that there is no image to be performed; and a fifth means for determining that an M × N area near the corresponding position in the input image is in the second position when it is determined that the recognition target exists at a certain position .
P '× Q' in small block corresponding to block size
, A representative value of each block is calculated, an arbitrary M ′ × N ′ region is taken out of the block, a distance between the region and the representative value group obtained by the second means is calculated, and the distance is calculated. The sixth means and the sixth means are repeatedly applied over the entire area of P ′ × Q ′, and the minimum value is obtained from the obtained distances, and the position and size of the M ′ × N ′ image at this time are determined. A seventh means for determining the position and size of the image to be recognized, and an area of the corresponding position and size in the input image obtained by the seventh means being reduced to a smaller block size corresponding to the third block size by M A representative value of each block is calculated by dividing into “× N”, and a distance between the representative value group obtained by the third means and a representative value group is calculated. An image recognizing device comprising an eighth means for performing the following. 2. The method according to claim 1, wherein the target image is not limited to a unique image group, but is a group of a plurality of similar images .
2. The image recognition apparatus according to claim 1 , wherein a representative value of each block is calculated , an average value of a plurality of blocks corresponding to a plurality of images is calculated, and the average value is stored as a dictionary. . 3. A said second means, not only the image to be pre-interest, a plurality of image groups to be similar, with each block is divided into M '× N' for each of these images
2. The image according to claim 1 , wherein a representative value is calculated for each of all the blocks , an average value among a plurality of blocks for a corresponding block between the plurality of images is calculated , and the average value is stored as a dictionary. Recognition device. 4. A said third means, not only the image to be pre-interest, a plurality of image groups, each blanking for each divided into M "× N" for each of these image blocks
2. The image recognition apparatus according to claim 1 , wherein a representative value for each lock is calculated and stored as a dictionary. 5. The method according to claim 2, wherein the partial image of the image to be recognized is divided into M ′ × N ′ in advance with a small block size corresponding to a second block size. Is calculated and stored as dictionary data.
And the seventh means, the partial area of the M × N area obtained by the fifth means is divided into P ′ × Q ′, and each of the M ′ × N ′ is divided into The image recognition device according to claim 1, wherein the distance is calculated. 6. The method according to claim 3, wherein the partial image of the recognition target image is divided in advance into M ″ × N ″ with a smaller block size corresponding to the third block size. Is calculated and stored as dictionary data. In the eighth means, the M ′ × N ′ area is stored in the third
2. The image recognition apparatus according to claim 1, wherein the image data is divided into M ".times.N" with a smaller block size corresponding to the block size. 7. The apparatus according to claim 1, wherein in the fourth means, a block size used for division is variable.
An image recognition device according to claim 1. 8. The method according to claim 1, wherein the block size used for the division is variable.
An image recognition device according to claim 1.