JPH0719277B2 - Image recognition device - Google Patents

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image recognition device, and more particularly to an image recognition device that accurately recognizes an image such as a character pattern by using rank correlation.

[Conventional technology]

An image recognition device for recognizing an image such as a character pattern inputs not only handwritten characters but also type or print characters, and various other image information to a mechanical device such as a computer without human intervention. It is a useful device that can be used for various purposes, and various types have been researched and developed.

[Problems to be Solved by the Invention] Conventionally, although various image recognition devices have been developed, those having complete recognition ability have not yet been developed, and more development is required. It is demanded to develop what has more excellent recognition ability based on the development of.

An object of the present invention is to provide an image recognition device capable of accurately recognizing an image by a relatively simple method.

[Means for solving problems]

The image recognition apparatus of the present invention includes a region defining unit that defines a region of a predetermined shape surrounding an image to be recognized, and a dividing unit that divides the region of the predetermined shape defined by the region defining unit into a plurality of predetermined sub-regions. A pixel number calculating means for obtaining the number of pixels of an image existing in each sub-region divided by the dividing means, and the plurality of sub-regions in descending order of the number of pixels of each sub-region obtained by the pixel number calculating means. And a rank value calculation means for calculating a rank value for each sub-area, and a rank value of each sub-area calculated by the rank-value calculation means is compared with a rank value of each sub-area of the reference image to rank the both. It has a rank correlation calculating means for obtaining the correlation, and a judging means for judging whether or not the images to be recognized are the same as the reference image based on the rank correlation calculated by the rank correlation calculating means.

[Action]

In the image recognition apparatus of the present invention, an area of a predetermined shape surrounding an image to be recognized is divided into a plurality of sub-areas, the number of pixels of the image existing in each of the divided sub-areas is calculated, and the number of pixels in descending order A plurality of sub-regions are ranked to obtain a rank value for each sub-region, this rank value is compared with the rank value of each sub-region of the reference image to obtain a rank correlation between the two, and recognition is performed based on this rank correlation. It is determined whether the power image is the same as the reference image.

〔Example〕

 Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a flowchart showing the operation of the image recognition apparatus according to the embodiment of the present invention. The image recognition apparatus of this embodiment shows a case of identifying a character pattern as image information, divides this character pattern into a plurality of sub-regions, and determines the rank value of the number of pixels in each sub-region as a reference image pattern, that is, a dictionary. The rank correlation between the two is calculated by comparing with the rank value of the pattern, and the character pattern is recognized by this rank correlation.

That is, if a character pattern "alternative" as shown in FIG. 1 (a) is input, a minimum rectangle including or surrounding this character pattern is calculated as shown in (b),
Further, this rectangle is divided into a plurality of sub-regions having the same size, that is, a plurality of M × N blocks as shown in FIG. 1 (c) (steps 110 and 120). In addition, in FIG. 1C, it is divided into 16 blocks of 4 × 4.

Next, the number of pixels existing in each block divided into M × N, that is, the number of black pixels in the illustrated case is obtained. In FIG. 1 (d), the number of pixels is shown by the number in each block divided into 16 pieces, and this number of pixels corresponds to the character “dai” shown in FIG. 1 (c). It is a thing. Then, the blocks are ranked in descending order of the number of pixels in each block thus obtained (step 130). FIG. 1 (e) shows that the numbers are assigned rank values from 1 to 16 in descending order based on the number of pixels shown in FIG. 1 (d). A block having a large number is given a rank value of 1, and a block having the smallest number of pixels of "5" is given a rank value of 16. The rank value ij is greater than or equal to 1 and less than or equal to M × N, that is, 1 ≦ ij ≦ M × N, where j = 1, 2, ..., M
× N.

In the next step 140, the rank value ij of the input character pattern obtained as described above is compared with the rank value αj of the reference image pattern, that is, the dictionary pattern, and the rank correlation γi _{α of the} two.
The dictionary configuration of the character data storing the rank value αj of the dictionary pattern will be described with reference to FIG.

In FIG. 2, the character pattern “substitution” is divided into blocks M × N, which are the minimum rectangle and a plurality of sub-regions, as in steps 120 and 130 described above, and the number of pixels contained in each block is calculated. Each block is ranked in descending order of number.

FIG. 2 (f) shows the rank values ranked in this way. In this way, the rank values αj for all character patterns are obtained, and these rank values αj are assigned to each character pattern. Correspondingly, the dictionary data is constructed by storing it in a memory or the like.

The rank value αj of the dictionary pattern thus configured is sequentially read from the memory or the like, the rank value αj of this dictionary pattern is compared with the rank value ij of the input character pattern calculated as described above, and the rank correlation of both patterns is compared. γ j _α is calculated (step 140). The rank value ij and the rank value αj are assigned in each block as shown in (f) and (g) of FIG. 1, respectively. The rank correlation γi _α of both patterns thus assigned is −1 ≦ γi _α ≦ 1,
Here, α = 1, 2, ... K.

After obtaining the rank correlation γi _α of both patterns, the maximum value γα _* of this rank correlation γi _α is obtained (step 150).

γ _α ^* = ma × {γ i _α | α = 1,2, ..., k} Then, the maximum value γ _α ^* of this rank correlation γ i _α is obtained for all dictionary patterns, and this maximum value γ _α ^* the largest dictionary pattern alpha ^* images, that is, to recognize as a character pattern corresponding to the character pattern in the input character pattern (step 160).

Next, rank correlation, particularly Spearman rank correlation, will be described.

The rank correlation is a two-rank value ij, which is ranked as shown in the following table by ranking each n characteristic in some two solids i and α having n characteristics by some criterion.
Find α j. The rank values ij and αj are 1 ≦ ij ≦ n, 1 ≦ α
j ≦ n, where j = 1, 2, ..., N.

The rank correlation γi _α is calculated from the rank values ij and αj thus calculated by the following equation.

Although the minimum rectangle including the character pattern is obtained and divided in the above embodiment, this shape is not limited to the rectangle and may be another shape including the image pattern to be recognized.

Further, the number of pixels calculated in each block is not limited to the number of black pixels, and conversely white or no pixels without black pixels or the equivalent may be used.

〔The invention's effect〕

As described above, according to the present invention, an image to be recognized is divided into a plurality of sub-regions, and a rank correlation between the rank value of the number of pixels included in each sub-region and the rank value of each sub-region of the reference image. Since the image is recognized on the basis of, the subdivided areas formed by dividing the image can also accurately recognize a relatively fine portion.

[Brief description of drawings]

FIG. 1 is a flow chart showing the operation of the image recognition apparatus according to one embodiment of the present invention, and FIG. 2 is an explanatory diagram of rank correlation used in the embodiment of FIG. In the figure, 110 to 160 ... Steps.

Claims (1)

[Claims] 1. A region defining means for defining a region of a predetermined shape surrounding an image to be recognized, and a dividing means for dividing the region of the predetermined shape defined by the region defining means into a plurality of predetermined sub-regions. Pixel number calculating means for obtaining the number of pixels of an image existing in each sub-region divided by the dividing means, and ranking the plurality of sub-regions in descending order of the number of pixels of each sub-region obtained by the pixel number calculating means Then, a rank value calculating means for obtaining a rank value for each sub-area, and a rank value for each sub-area obtained by the rank-value calculating means is compared with a rank value for each sub-area of the reference image to obtain a rank correlation between them. An image recognition apparatus comprising: a rank correlation calculating means; and a judging means for judging whether or not the image to be recognized is the same as the reference image based on the rank correlation calculated by the rank correlation calculating means. .