JP2855665B2 - Character recognition method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a character recognition system, and more particularly to a character recognition system having a small storage capacity and a small amount of arithmetic processing.

(Prior art) With the diversification of information processing systems, various data input methods are required, and character recognition is being put to practical use as a powerful data input method. At the same time, with the aim of improving recognition performance. Various character recognition schemes have been proposed.

One of them is a previous application by the same applicant (Japanese Patent Application
-237066). This proposal is characterized by dividing a character pattern into a plurality of sub-regions, and obtaining a two-dimensional vector sequence in each of the directions in which the opposing directions on the same direction are paired in a plurality of direction components. This is a character recognition method that describes a pattern and applies nonlinear matching to each two-dimensional vector sequence.

FIG. 2 shows an example of a scalar sequence of directional components in consideration of the orientation of a partial region of a character pattern. A two-dimensional vector sequence having a pair of opposite directions in the same direction is f
Assuming that k of (i, k) is an index indicating a direction and an index indicating a series, for example, it is expressed as follows.

(F (1,4), f (1,8)), (f (2,4), f (2,8)),
(F (3,4), f (3,8)),... (Problems to be Solved by the Invention) In the above-described conventional character recognition method, the feature description and the calculation of the degree of difference are executed in units of two-dimensional vectors. for,
There is a problem that the storage capacity for storing the reference pattern of each character type to be read is large, and the processing amount of the difference calculation is large.

(Means for Solving the Problems) In order to solve the above-mentioned problems, the character recognition method of the present invention comprises:
A character pattern storage unit for storing a character pattern given as a two-dimensional lattice array; reading a character pattern signal from the character pattern storage unit; A direction pattern corresponding to a plurality of directions is obtained by obtaining a direction component distinguished by a code, and a scalar is obtained for each direction component in a partial area on each direction pattern corresponding to a plurality of divided partial areas on the character pattern. A feature extraction unit that performs feature description in a series of quantities to create a feature pattern; a feature pattern storage unit that stores a feature pattern obtained by the feature extraction unit; a feature stored in the feature pattern storage unit A reference pattern storage unit storing a reference pattern for each character type to be read in the same format as the pattern; The characteristic pattern is read from the signature pattern storage unit, the reference pattern for each character type to be read is read from the reference pattern storage unit, and a series of scalar amounts of each partial region of the characteristic pattern and a series of the scalar amount are obtained. By performing non-linear matching between the series of scalar amounts of the corresponding reference patterns, the degree of difference between each of the corresponding partial areas of the feature pattern and each of the reference patterns is calculated. A character recognition method including a classification processing unit that calculates a difference between the feature pattern and each of the reference patterns, ranks the differences, and outputs a classification result.

(Operation) In the present invention, for a character pattern in a two-dimensional lattice arrangement, a direction pattern in which opposite directions in the same direction are distinguished by signs of positive and negative values is obtained, and a feature description is performed using a scalar quantity sequence for each direction component. The non-linear switching is performed between the feature pattern and the reference pattern of the same format to calculate the difference between the two patterns. Therefore, according to the present invention, the storage capacity and the amount of arithmetic processing are significantly reduced as compared with the related art.

(Example) Next, the present invention will be described with reference to the drawings.

First, before describing embodiments of the present invention, the principle of the present invention will be described with reference to the drawings.

FIG. 2 shows a characteristic sequence f (i, k) (i = 1,..., I; k =
1, 8) are shown. The directions of the feature series indicated by the dependent variables i of the feature series f (i, k) and f (i, k + 4) (k = 1,2,3,4) in the same direction are as shown by arrows in the figure. The same shall apply. In the previously proposed character recognition method,
A set of two feature sequences in opposite directions is defined as a feature sequence related to the direction k as a two-dimensional vector (f (i, k), f (i, k + 4)) (k = 1,2,3,4). .

FIG. 3 (a) shows a two-dimensional vector (f (i, k), f (i, k
+4)) (k = 1, 2, 3, 4): This is a diagram showing an example of a sequence. As shown in this figure, a sequence f (i, k) and a sequence f (i, k + 4)
If the peaks are sufficiently separated, as shown in FIG. 3 (b), the sequence f (i, k) and the sequence {−f
(I, k) (= f (i, k, 4))
k) -f (i, k + 4)) two original sequences f (i, k) and f
(I, k + 4) information can be stored. When the sequence (i, k) thus synthesized is applied as a feature sequence, (1) the storage capacity for the reference pattern is halved compared to when a two-dimensional vector sequence is applied, and (2) when performing nonlinear matching This has the advantage that the amount of calculation is reduced by half.

Sequence (i, k) obtained by combining as described above
Is a scalar series. As a nonlinear matching method applied to such a scalar sequence, for example, literatures (Journal of the Acoustical Society of Japan, 27, 9, pp. 483-500 (1976)), Sakoe, Chiba,
It has been shown that the DP matching method proposed in "Continuous word recognition based on temporal normalization of speech using dynamic programming") is effective. The degree of difference between the two character patterns F and G is obtained by the following calculation method by applying the above-described feature sequence and DP matching. In the following equation, (i, k; m, n) (i = 1,..., I, k = 1,.
= 1,..., N) indicates a feature sequence in the partial region (m, n) of the character pattern F, and (i, k; m, n) indicates a feature sequence in the partial region (m, n) of the character pattern G. I have.

In the above expression, D (k; m, n) is the degree of difference in the pattern of the scalar series in the direction k in the partial area (m, n), and the degree of difference D (m, n) in the partial area is The degree of difference between patterns is determined by a calculation method shown in the following equation (4).

The dissimilarity D of the entire pattern is obtained by a calculation method for calculating the dissimilarity of each partial area as shown in the following equation (5).

By performing the above calculations, the degree of difference between the input character pattern and the reference pattern for each character type to be read is obtained, and the ordering is performed in ascending order of the degree of difference, whereby the classification process can be realized.

 FIG. 1 is a block diagram showing one embodiment of the present invention.

The character pattern storage unit 1 stores an input character pattern, and ordinary storage means is used.

The feature extraction unit 2 reads a character pattern signal from the character pattern storage unit 1 and extracts a feature of each partial region determined by dividing the character pattern as a series having a positive or zero value in a plurality of directions in consideration of the orientation. Then, a series of two scalar quantities in opposite directions in the same direction is described as a series synthesized by subtraction, and is output as a characteristic pattern signal. The details will be described later.

The feature pattern storage unit 3 stores the feature pattern signal output from the feature extraction unit 2, and is configured by ordinary storage means.

The reference pattern storage unit 4 stores, as a reference pattern, a feature pattern for each read target character type described in the same format as the feature pattern storage unit 3, and may be a normal storage unit.

The classification processing unit 5 reads the characteristic pattern signal from the characteristic pattern storage unit 3, reads the reference pattern signal for each character type to be read from the reference pattern storage unit 4, and as described above, the scalar for each direction in each partial area. The degree of difference of the feature represented as a series is calculated by DP matching, the degree of difference of each partial area is calculated by the sum of the degree of difference of each direction, and the input character pattern and each reference pattern are calculated by the sum of the degree of difference of each partial area And a classification result is determined by ordering the degrees of difference, and a classification result signal is output. This can be easily realized by the technology applied to the conventional pattern recognition device.

FIG. 4 is a block diagram showing an example of the configuration of the feature extracting unit 2.

The directional distribution extraction unit 21 reads the character pattern signal from the character pattern storage unit 1, extracts the distribution of the directional component for each direction in consideration of the direction, and extracts the directional distribution pattern signal of each direction in consideration of the direction. It can be easily realized by conventional pattern recognition technology and image processing technology.

The directional distribution pattern storage unit 22 stores the directional distribution pattern output from the directional distribution extraction unit 21, and ordinary storage means may be used.

The projection area storage unit 23 stores each position coordinate of each projection area for performing a projection process for obtaining a sequence in each direction in consideration of the direction of each partial area, and may be a normal storage unit.

The partial projection extraction unit 24 reads the position coordinate signal of each projection area in each direction in consideration of the orientation in each partial area from the projection area storage unit 23, and stores the corresponding directional distribution pattern signal in the directional distribution pattern storage unit. The scalar sequence signal is output in each direction in consideration of the direction of each partial area by reading from the subroutine 22 and obtaining each partial projection, and can be realized by conventional pattern recognition technology and image processing technology.

The feature description unit 25 reads a scalar sequence signal in each direction in consideration of the direction output from the partial projection extraction unit 24, and, for each direction of each partial region, two scalar sequences corresponding to opposite directions on the same direction. Is described as a sequence synthesized by subtraction as described above to create a feature pattern and output it as a feature pattern signal, and can be easily realized by conventional techniques such as digital circuit technology.

(Effects of the Invention) As described above, according to the present invention, in each partial region on a character pattern, a feature sequence corresponding to an opposite direction for each direction is treated as a scalar sequence, and nonlinear matching is applied. A highly accurate character recognition method that can locally correspond to an input character pattern and a reference pattern of each read character type can be realized with a compact storage capacity and efficient difference calculation as compared with the related art.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the configuration of the present invention, FIG. 2 is a diagram showing an example of a feature sequence in a plurality of directions in consideration of the orientation to a pattern in a partial area, and FIG. FIG. 4 is a diagram showing a scalar sequence obtained for an orientation and a scalar sequence obtained by synthesizing the scalar sequence. FIG. 1. Character pattern storage unit 2. Feature extraction unit 3.
Feature pattern storage unit, 4 ... Reference pattern storage unit, 5 ...
... Classification processing unit, 21 ... Directed direction distribution extraction unit, 22 ... Directed direction distribution pattern storage unit, 23 ... Projection area storage unit, 24 ...
... partial projection extraction unit, 25 ... characteristic description unit.

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06K 9/46 G06K 9/62 630

Claims (1)

(57) [Claims] A character pattern storage unit for storing a character pattern given as a two-dimensional lattice array; reading a character pattern signal from the character pattern storage unit; A direction pattern corresponding to a plurality of directions is obtained by obtaining a direction component in which the direction is distinguished by a sign of a positive / negative value, and a partial pattern on each direction pattern corresponding to a plurality of divided partial areas on the character pattern is obtained. A feature extraction unit that creates a feature pattern in which features are described in a series of scalar quantities for each direction component; a feature pattern storage unit that stores a feature pattern obtained by the feature extraction unit; A reference pattern storing a reference pattern for each character type to be read in the same format as the stored characteristic pattern Reading the feature pattern from the feature pattern storage unit, reading the reference pattern for each character type to be read from the reference pattern storage unit, and obtaining a series of scalar amounts of each partial region of the feature pattern; By performing non-linear matching between the series of scalar amounts of each reference pattern corresponding to the series of scalar amounts, the degree of difference between each of the partial regions corresponding to the feature pattern and each of the reference patterns is calculated. A character recognition method, comprising: a classification processing unit that calculates a degree of difference between the feature pattern and each of the reference patterns from a degree of difference in a region, ranks the degrees of difference, and outputs a classification result.