JP2755738B2 - Character recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (a) Field of Industrial Application The present invention relates to a character recognition device, which minimizes recognition candidate characters in order to simplify correction work and improve post-processing accuracy such as word matching. The present invention relates to a character recognition device that narrows down the search.

(B) Conventional technology In conventional character recognition, as a recognition result, there are some candidate characters that are fixed to a fixed number (usually about 10). The Institute of Electronics, Information and Communication Engineers (D) Vol.J71-D No.6 pp10
37-1047 As disclosed in a paper "Character Pattern Extraction and Recognition from Scene Images" in June 1988, a threshold value of the distance (or similarity) is set for each character category, and the number is variable. There are things. Furthermore, it is disclosed in the IEICE Transactions D-II Vol.J72-D-II No.7 pp993-1000 in July 1989, "Error Correction Processing Method for Kanji Recognition by Candidate Character Completion and Language Processing". As described above, although the number of candidate characters was narrowed down to some extent by various methods such as those limited by a fixed threshold value, it was not sufficient.

(C) Problems to be Solved by the Invention It is desirable that the number of candidate characters be as small as possible in view of the complexity of the selecting side.

In each of the conventional methods, candidates are not sufficiently narrowed down, and in the correction work, characters that are hardly similar are listed as candidates at the time of candidate selection, and selection is complicated.

In addition, when performing post-processing such as word matching after recognition, the number of candidate characters increases more than necessary, and the number of combinations increases. was there.

SUMMARY OF THE INVENTION In view of the above-described conventional problems, an object of the present invention is to narrow down the number of candidate characters without deteriorating recognition accuracy, to facilitate correction work, or to increase post-processing accuracy.

(D) Means for Solving the Problems The present invention provides a standard pattern dictionary having a standard font for each character category, and a standard pattern between an input character pattern and a character type registered in the standard pattern dictionary. An inter-pattern distance calculating unit for calculating an inter-distance, a distance order rearranging unit for rearranging characters in ascending order of the calculated distance, a first candidate character narrowing unit for roughly limiting candidate characters, In addition to calculating the distance difference between adjacent characters, the distance difference calculator and the average and variance of the distance difference are calculated, and it is determined whether or not the candidate characters can be narrowed down. And a second candidate character narrowing unit that divides the characters into characters, and by finding a position where the distance difference becomes the maximum, leaving characters having a smaller distance as candidate characters, and rejecting others. And

(E) Function Whether or not the candidate characters can be narrowed down is determined by whether or not the variation in the distance difference is large or not. A distance difference occurs. Therefore, the separation point between the candidate character and the rejected character is obtained for the distance difference, and the candidates are narrowed down. On the other hand, when there are many similar characters, it is not necessary to forcibly narrow down the candidates, and all the candidate characters are left.

As described above, according to the present invention, candidate characters are reasonably and effectively narrowed down.

(F) Example Hereinafter, an example of the present invention will be described with reference to the drawings.

In the present invention, the narrowing down of candidate characters is performed by arranging candidate characters in the order of smaller distance (in order of larger similarity).
This is performed based on the distance difference.

First, the concept of the present invention will be described with reference to FIG. First
FIG. 5 is a conceptual diagram showing a state in which candidate characters are arranged in ascending order of distance.
It is made into pieces. In the figure, a black circle represents one candidate character, and the distribution in the distance axis direction is shown with the distance as the horizontal axis.

FIG. 1A shows a case where there is a variation in the distance difference,
The distance difference is maximum at the location of the broken line in the figure. That is, there is a significant distance difference between the third character and the fourth character, and the character to be included in the candidate characters and the group of characters to be rejected can be separated from the difference.

FIG. 1 (b) shows a case where the distance difference is almost the same. In this case, separation into two groups as in the case of FIG. 1 (a) is dangerous. In this case, forcibly narrowing down candidates is avoided, and all characters are left as candidate characters.

When applied to an actual case, FIG. 1A corresponds to a case where there are relatively few similar characters, and FIG. 1B corresponds to a case where there are many similar characters.

In the present invention, first, it is necessary to determine whether this is the case in FIG. 1 (a) or (b).

To explain this determination, first, the average of the distance differences between the candidate characters is obtained, and whether or not the average is extremely larger than that value is used as a criterion. That is, if there is an extremely large one, it is considered that the candidate characters can be further narrowed down. If not, no further refinement is performed.

In this method, the number of candidate characters needs to be limited to a few characters in advance (10 candidates in FIG. 1).
For this reason, as a first step, the number of candidate characters is limited to a few using the conventional method. In the second stage, the distance difference between the candidate character groups is obtained by the above-described method, and if possible, the characters are divided into characters to be included in the candidates and characters to be rejected.

Next, the present invention will be further described with reference to FIGS.

FIG. 2 is a block diagram showing an example of the configuration of the present invention, and FIG. 3 shows a processing flow.

The inter-pattern distance calculation unit 2 performs a feature comparison on the input character pattern 1 and all the character types registered in the standard pattern dictionary 3 to calculate an inter-pattern distance from the standard pattern. The standard pattern dictionary 3 stores standard fonts for each character category. Pattern distance calculation unit 2
, That is, the character type and its distance value are provided to the distance order rearranging unit 4. In the distance order rearranging section 4, as shown in FIG. 1, the distances are rearranged in ascending order. The rearranged data is provided to the primary candidate character narrowing unit 5. The primary candidate character narrowing unit 5 narrows down to several candidates based on the given data. In this process,
Broadly narrow down candidates using conventional methods. In this process, the number of candidate characters may be simply reduced to a constant number (for example, about 10), or the number of candidates may be reduced by setting a threshold value for each character category.

Subsequently, the data is sent to the second candidate character narrowing section 6. The second candidate character narrowing unit 6 includes a distance difference calculating unit 61, an average and variance calculating unit 62, a narrowing down possibility determining unit, and a candidate character separating unit 64, and performs the following processing.

First, if necessary, the distance values are rearranged in ascending order, and are omitted in FIG. 2 (in the present embodiment, they are already arranged in the order of distance at the time of the primary candidate character narrowing unit). The data is given to the calculation unit 61. The distance difference calculator 61 calculates the difference between the distances of adjacent characters. The average and variance calculator 62 calculates the average and variance of the distance difference. Then, the narrowing down determination unit 63 determines whether or not there is an extremely large distance difference based on the average value and the variance. Specifically, if there is a distance difference satisfying the following equation, the candidate character separation unit 64
However, the candidate is divided at the maximum distance difference, and the candidate character and the character candidate at a distance longer than that are rejected.

(Distance difference)> (average distance difference) + (twice the variance of the distance difference) If there is no distance difference that satisfies the above equation, the characters selected by the primary candidate character narrowing unit 5 Is adopted as a candidate character as it is.

The obtained candidate characters are passed to the post-processing unit 7, where processing such as word matching is performed.

Although not shown in the example of FIG. 2, when post-processing is not performed, the character is adopted as the next candidate character at the time of correction.

Next, the processing procedure of the present invention will be described with reference to the flowchart of FIG.

In step S 1, the inter-pattern distance calculation unit 2 performs a feature comparison on the input character pattern and all the character types registered in the standard pattern dictionary 3.

Subsequently, in step S2, the result obtained in step S1 (character type and its distance value) is
, And are sorted in ascending order of distance.

In step S3, the first candidate character narrowing unit 5 narrows down to several candidates.

Thereafter, the processing in the second candidate character narrowing section 6 is performed. The following processing is performed in the candidate character narrowing section 6 in FIG.

In step S4, the distance difference calculation unit 61 calculates the difference in distance between adjacent characters. Then, in step S5, the average and variance calculation unit 62 calculates the average and variance of the distance difference, and then proceeds to step S6.

In step 6, based on the average value and the variance, the narrowing down determination unit 63 determines whether or not there is an extremely large distance difference.

If there is the above-mentioned one having the predetermined distance difference, the process proceeds to step S7. In step S7, the candidate character separation unit 64
Splitting is performed at the maximum distance difference, character candidates at distances longer than that are rejected, and the operation ends.

On the other hand, if there is no difference in distance, the process proceeds to step S8, the character selected in the primary candidate character narrowing process is adopted as a candidate character as it is, and the operation is terminated.

(G) Effects of the Invention As described above, according to the present invention, candidate characters can be significantly narrowed without impairing the character recognition accuracy, and the complexity of user selection of candidate characters during correction can be reduced. it can. Also in the case of performing post-recognition processing, by reducing the number of candidate characters, the number of combinations of word matching is reduced, and the processing can be speeded up and the accuracy can be increased.

[Brief description of the drawings]

FIG. 1 is a conceptual diagram of the present invention, FIG. 2 is a block diagram showing an embodiment of the present invention, and FIG. 3 is a flowchart of processing in the present invention. 2... Inter-pattern distance calculation unit, 3... Standard pattern dictionary, 4... Distance order rearranging unit, 5... Primary candidate character narrowing unit, 6.

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-3-71285 (JP, A) JP-A-53-80922 (JP, A) JP-A-61-72376 (JP, A) JP-A-61-72376 114299 (JP, A) JP-A-62-192890 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) G06K 9/62 Patent file (PATOLIS) JICST file (JOIS)

Claims (1)

(57) [Claims] 1. A standard pattern dictionary having a standard font for each character category, an inter-pattern distance calculation unit for calculating a standard inter-pattern distance between an input character pattern and a character type registered in the standard pattern dictionary. A distance ordering unit that sorts characters in ascending order of the distance calculated by the calculating unit; a primary candidate character narrowing unit that roughly limits candidate characters; a distance difference between adjacent characters after sorting Divide the candidate character and the rest at the maximum distance difference, and the distance difference calculation unit, the average and variance calculation unit that calculates the average and variance of the distance difference, the narrowing down determination unit that determines whether narrowing down the candidate characters A candidate character separating unit comprising: a candidate character separating unit; and a second candidate character narrowing unit comprising: a candidate character separating unit; Left as characters, character recognition apparatus characterized by otherwise rejected.