JPH05135205A - Character recognition device - Google Patents

Abstract

PURPOSE:To improve the recognition precision by attaining a multilevel discrimination processing without increasing a circuit scale. CONSTITUTION:The feature pattern of a character picture is extracted as a multi-dimensional vector by a feature extracting circuit 31, and compared with a dictionary pattern stored in a storage device 33 for a dictionary by a pattern matching circuit 35 in order to search the similarity. And also, a code indicating a discrimination level is added to the similarity, and the obtained similarity data are rearranged in the order of the similarity by a sort circuit 39 in order to operate a discrimination processing. Moreover, the discrimination processing of the feature pattern extracted from the character picture by a different method is repeated by using the same pattern matching circuit 35 and circuit 39.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device, and more particularly to a character recognition device for performing discrimination processing by multi-step pattern matching.

[0002]

2. Description of the Related Art The processing of a character recognition device using a general pattern matching method is performed as shown in the flowchart of FIG. First, an image is cut out from the input document image character by character. Next, a specific image conversion is performed on this character image to extract a characteristic pattern.
Here, the feature of the pattern obtained by the image conversion used is referred to the dictionary part stored in advance for each character type, and compared with the dictionary pattern by sequential pattern matching to find the most similar. Outputs the existing character type as the recognition result.

This character type discrimination processing is performed in a plurality of stages in order to reduce the amount of calculation and improve the discrimination accuracy. That is,
The candidates are narrowed down by first performing rough pattern matching processing and then performing more detailed pattern matching processing. In this flowchart, the character type is discriminated by two steps of classification processing and recognition processing. When the character type is discriminated in a plurality of steps, the recognition accuracy is improved by using patterns extracted by different methods for each discriminating process.

FIG. 6 is a block diagram showing a circuit configuration for performing conventionally used character type discrimination. As shown in the figure, the classification processing unit 1 includes a classification feature storage device 11 that stores a feature pattern extracted by the feature extraction circuit 3 from a character image to be recognized, and a classification dictionary storage that stores a dictionary pattern. The device 12, a pattern matching circuit 13 that sequentially compares the dictionary pattern of the storage device 12 with the recognition target feature pattern stored in the storage device 11 to obtain the degree of similarity, and character code generation for the compared dictionary pattern. The sorting circuit 14 arranges the character codes sent from the circuit 4 in the order of similarity.

The results of the classification processing by the classification processing unit 1 are output as character codes of a plurality of candidate characters together with the order of similarity, and are temporarily stored in the work memory 5, and then the recognition processing unit 2 and the address generation circuit 6 are provided. Sent to. The address generation circuit 6 generates an address of the corresponding recognition processing storage device 22 from each character code of the candidate character and sends it to the recognition processing unit 2. The recognition processing unit 2 is configured almost the same as the classification processing unit 1, executes the same discrimination processing, and selects and outputs the character code with the highest similarity as the processing result. The character code in the work memory 5 is directly input to the sort circuit 24.

[0006]

By the way, in the above-mentioned conventional character recognition device, when an attempt is made to increase the number of stages of the character type discrimination processing in order to improve the recognition rate and reduce the amount of calculation, the discrimination circuits corresponding to the increased number of stages are used. However, there is a problem that the circuit scale becomes large. These circuits are
Even when the SIC is used, several thousand gates are required for each work memory for passing the character candidate code, which is inefficient. The present invention has been made to solve the above problems, and an object of the present invention is to provide a highly accurate character recognition device having a relatively small circuit configuration.

[0007]

In order to achieve the above-mentioned object, a first invention is that an optically input character image
Pattern extracting means for extracting a plurality of types of characteristic patterns using mutually different methods, a dictionary unit in which the characteristic patterns obtained for each of the different extraction methods are stored in advance with a character code, and the characteristic patterns of the dictionary unit are sequentially called and extracted. A pattern matching circuit comprising a pattern matching circuit for obtaining a similarity by comparing with a characteristic pattern and a sort circuit for holding a character code having a high similarity and changing the order of the character codes according to the similarity. A means for adding a code indicating a discrimination step to the degree of similarity determined in step S1, and a plurality of types of character patterns obtained from the same character image are repeatedly used by a pair of pattern matching circuits and sort circuits to perform character discrimination. And a means for performing.

A second invention is characterized in that in the first invention, there is provided means for expressing the extracted feature pattern as a multidimensional vector.

[0009]

In the present invention, the characteristic pattern of the character image extracted by the pattern extracting means is represented as a multidimensional vector. In addition, a code indicating a discrimination step is added to the similarity obtained by the pattern matching circuit, so that the discrimination step can be identified. Furthermore, a pair of pattern matching circuits and sort circuits are repeatedly used to perform multistage character determination for each type of extracted feature pattern.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a main part of a character recognition device to which the present invention is applied. In the figure, 31 is a feature extraction circuit, which extracts a feature pattern from a character image to be recognized by using a different method for each discrimination step, represents the feature pattern as a multidimensional vector, and sends it to the feature storage device 34. .. The address generation circuit 32 generates an address for referring to the dictionary pattern based on the discrimination stage identification code and the candidate character code, and sends it to the dictionary storage device 33 and the characteristic storage device 34. In particular, when the character image is first discriminated, the address generation circuit 32 sequentially generates addresses for all the character types in the dictionary storage device 33.

The dictionary storage device 33 as a dictionary unit stores typical characteristic patterns for each character type as dictionary patterns for each discrimination step, and when an address is designated, the corresponding dictionary pattern is multidimensional. It is sent to the pattern matching circuit 35 as a vector. In this embodiment, by storing the dictionary pattern for each discrimination stage in the dictionary storage device 33, the storage device can be downsized. The characteristic storage device 34 temporarily stores the characteristic pattern of the character image to be recognized as a multidimensional vector for each discrimination step, and sends the characteristic pattern to the pattern matching circuit 35 in the discrimination step order.

The pattern matching circuit 35 compares the feature pattern input as a multidimensional vector with the dictionary patterns sequentially input, calculates the degree of similarity by an algorithm described later, and sends it to the sorting circuit 39. The discrimination stage identification circuit 36 sends the stage of the discrimination process being performed at that time to the address generation circuit 32, the selector 38 and the sorting circuit 39 as a discrimination stage identification code. The character code generation circuit 37 generates a character code of the compared dictionary pattern and sends it to the selector 38 every time the similarity is sent from the pattern matching circuit 35 to the sort circuit 39.

The selector 38 sends the character code output from the character code generation circuit 37 to the sort circuit 39 if the content of the input discrimination stage identification code is the first discrimination processing, and the content of the discrimination stage identification code is the first time. If not, the character code of the candidate character output from the sorting circuit 39 is sent to the sorting circuit 39. The sorting circuit 39 arranges the dictionary patterns having a high degree of similarity in descending order of the degree of similarity, holds the character codes of the upper N candidate characters used in the discrimination process in the next stage, and stores the character codes in the address generating circuit 32. , Selector 38 and the outside.

Next, the process of obtaining the degree of similarity in the pattern matching circuit 35 will be described. The similarity is 2
It is a numerical value indicating the degree of matching between two patterns, and the higher the degree of similarity, the more similar the two patterns are. In the present invention, the similarity Sj is obtained from the equation 1 by considering the feature pattern used in each discrimination step as a vector in the multidimensional space.

[0015]

[Equation 1]

In the equation, | A | and | B | represent the absolute value of the vector, and (AB) represents the inner product of the vectors A and B. Here, A = (a1, a2, ... An), B =
If (b1, b2, ... bn), then the number 2,
The relationships of Equations 3 and 4 are obtained.

[0017]

[Equation 2]

[0018]

[Equation 3]

[0019]

[Equation 4]

When the characteristic pattern extracted from the input character image is A and the dictionary pattern of the character type J is Bj,
The pattern matching circuit 35 repeats the operation of Expression 1 for the dictionary patterns Bj of all candidate characters. Where |
A | is constant during one discrimination process, and | Bj |
Is already known, so if Gj = 1 / | Bj | is written in the dictionary storage device in advance, the actual discriminant is
It becomes like the number 5.

[0021]

[Equation 5]

From these results, the internal structure of the pattern matching circuit 35 is as shown in FIG. That is, in the product-sum calculator 41, ai and bi are input by the equation 4 and the inner product is obtained. Further, in the multiplier 42, the similarity Sj is obtained from Gj and the inner product from the product-sum calculator 41 according to Equation 5. In this way, when the feature pattern is represented as a multidimensional vector and the similarity is obtained by obtaining the inner product of them, the similarity can be obtained regardless of the number of dimensions of the vector. Therefore, when the degree of similarity is obtained in the actual multi-step discrimination process, the pattern matching circuit 35 can be shared even if the characteristic patterns to be compared for each discrimination step are represented as vectors having different numbers of dimensions. ..

Next, the sorting circuit 39 will be described.
FIG. 3 is a block diagram showing the configuration of the sort circuit 39.
FIG. 4 is a block diagram showing the internal configuration of the sorting unit 51 that constitutes the sorting circuit 39. This sort circuit 39
Is already filed by the present applicant (JP-A-62-237523). This sort circuit 39
Is N sort units 51, which is the same as the number of candidate characters.
The sorting unit 51 further comprises
As shown in FIG. 4, respectively, a latch circuit 52 for storing data,
It is composed of a comparator 53 for data comparison and a data selector 54.

Assuming that the sorting unit 51 is located, for example, Mth (1≤M≤N) from the top, the sorting unit 51 holds data whose similarity is Mth (similarity, character code, etc.). There is. When data is newly input into the sorting circuit 39, the similarities between the two are compared. If the similarity of the newly input data is larger than the data held until then, the Mth data is transferred to the lower unit (M + 1), and at the same time, the data transferred from the upper unit (M-1) is transferred. It is captured and held as Mth data. On the contrary, if the Mth data is larger, the data newly input to the sort circuit 39 is transferred to the lower unit while the Mth data is held. When all sort units 51 perform these operations at the same time, the data rearrangement operation of the entire sort circuit 39 is realized.

Further, the sorting circuit 39 needs to carry out the data rearranging operation for the current time while holding the result of the previous discrimination processing. For that purpose, the discrimination stage identification code is added to the data as the higher order bit of the similarity so that the data of different discrimination stages are not mixed with each other, and the discrimination code of that time is expressed as a numerical value larger than the previous discrimination code. Keep it. For example, when performing a four-step discrimination process, the identification code for the first time is set to "0", and the second and subsequent times are sequentially incremented to "1",
Set to "2" and "3". By doing so, the processing stage is identified for the data obtained by the discrimination process at each stage, and it becomes possible to prevent the data from being mixed.

Furthermore, if the lowest N character code held as a character candidate is output from the sorting circuit 39 and this is used as the character code of the character type for which the pattern matching processing is performed at that time, By inputting new data to the sorting circuit 39, the character code for the next process is pushed out and output, and it becomes easy to refer to the character candidates. As described above, in the embodiment, by expressing the characteristic patterns to be compared as a multidimensional vector, the character type can be discriminated by the similar similarity calculation which is different only in the number of dimensions, and the same pattern matching circuit 35 and sort circuit 39 are repeatedly used. Then, it is possible to perform the multistage determination process for each type of extraction pattern. Also,
By adding a code representing a discrimination step to the similarity obtained by the pattern matching circuit 35, it is possible to prevent data of different discrimination steps from being mixed together. As a result, even if the number of stages of character type discrimination processing is increased to improve the recognition rate and reduce the amount of calculation, there is no need to increase the circuit scale at all, and the circuit portion of the character recognition device can be significantly downsized. ..

[0027]

As described above, according to the present invention, the characteristic pattern of the character image extracted by the pattern extracting means is represented as a multidimensional vector, and the similarity degree obtained by the pattern matching circuit indicates the discrimination step. With the addition of, it becomes possible to repeatedly use a pair of pattern matching circuits and sort circuits to perform multistage character determination for each type of extraction pattern. As a result, highly accurate character recognition is possible with the circuit kept in a relatively small size.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a main part of an embodiment of the present invention.

FIG. 2 is a block diagram showing an internal configuration of a pattern matching circuit.

FIG. 3 is a block diagram showing a configuration of a sorting circuit.

FIG. 4 is a block diagram showing an internal configuration of a sorting unit.

FIG. 5 is a flowchart showing the overall processing operation of the character recognition device to which the present invention is applied.

FIG. 6 is a block diagram showing a circuit configuration of a conventional example.

[Explanation of symbols]

 31 Feature Extraction Circuit 32 Address Generation Circuit 33 Dictionary Storage Device 34 Feature Storage Device 35 Pattern Matching Circuit 36 Discrimination Stage Discrimination Circuit 37 Character Code Generation Circuit 38 Selector 39 Sorting Circuit 41 Product Sum Operator 42 Multiplier 51 Sorting Unit 52 Latch Circuit 53 Comparator 54 Data selector

Front page continuation (72) Inventor Yasuo Hongo 1 Fuji-cho, Hino-shi, Tokyo Inside Fuji-Fuacom Control Co., Ltd.

Claims (2)

[Claims] 1. A pattern extracting means for extracting a plurality of types of characteristic patterns from optically input character images by using different methods, and a characteristic pattern obtained for each of the different extracting methods are stored in advance together with a character code. The dictionary part and the pattern matching circuit that calls the feature patterns of the dictionary part in order and calculates the similarity by comparing with the extracted feature pattern, holds the character code with a high similarity, and changes the order of the character code according to the similarity. In a character recognition device including a sorting circuit for replacement, a means for adding a code indicating a discrimination step to the similarity obtained by the pattern matching circuit, and a pair of plural types of character patterns obtained from the same character image. Repeatedly use the pattern matching circuit and the sort circuit of to perform multi-stage character recognition A character recognition device comprising means and. 2. The character recognition device according to claim 1, further comprising means for expressing the extracted feature pattern as a multidimensional vector.