JPH06309507A - Character recognizing device - Google Patents

Abstract

PURPOSE:To improve the recognizing factor for the input characters of different forms without increasing the number of dictionaries. CONSTITUTION:The input images are segmented the degrees of resemblance are decided among these images, and a candidate character string is produced (Steps 1-3). Then, it is decided whether the head character of the candidate character string is written or not in a column B of a table (Steps 4 and 5). The true codes of characters which are easily recognized wrong are written in a column A of the table, and the character codes obtained by recognizing wrong the codes written in the column A are written in the column B respectively. If the head character of the candidate character string is written in the column B, the corresponding character code or the character cluster code written in the column A is read. Then, a fact whether a code A is included in the candidate character string or not is discriminated (Steps 6 and 7). If the code A is included in the character string, the code A is shifted to the head of the character string. Thus, the code A is outputted as the result of recognition.

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 for cutting out an unknown character pattern from an optically input document image and recognizing it by using a dictionary pattern.

[0002]

2. Description of the Related Art In the conventional character recognition by pattern matching, the feature amount is extracted from the standard character image data for each character and stored as a dictionary, and the feature amount is extracted from the input unknown character image data. Is extracted and the correlation calculation is performed with the feature amount of each character in the dictionary, and the character having the largest correlation value (hereinafter referred to as similarity) is taken as the reading of the unknown character. Further, in order to improve the recognition accuracy, a plurality of dictionaries may be prepared for different fonts such as Mincho and Gothic.

[0003]

By the way, in the conventional character recognition, when an unknown character of a font which is not prepared as a dictionary is input, the similarity of each character in the dictionary with respect to the unknown character is reduced, but the entire dictionary is reduced. Since the degree of similarity of is uniformly decreased, it can be recognized correctly for the time being. However, if the input character and the font of the dictionary are different, the character shape may be similar depending on the character, so that the character may be erroneously recognized. For example, as shown in FIG. 9, an input character “U” having a different font from the dictionary may have a shape similar to the character “RO” in the dictionary. Therefore, if the similarity of the input character “U” with the dictionary is obtained, candidate character strings as shown in FIGS. 10 and 11 are obtained.

The candidate character strings are arranged in descending order of similarity, and although the correct reading "U" is included in the candidate character string in each case, the beginning is an incorrect reading "RO", so an incorrect recognition result is output. Will be done. As a result, when the character string shown in FIG. 2 is input, it is read as shown in FIG. 3 and erroneous recognition occurs at the end. Note that FIG. 11 shows a case in which a plurality of dictionaries are provided for the same character, and the number below the character represents the character cluster code. The present invention has been made to solve the above problems, and an object of the present invention is to prevent erroneous recognition when the font of an input unknown character and the font of a dictionary are different and improve the recognition accuracy. It is to provide a character recognition device capable of performing.

[0005]

In order to achieve the above object, the present invention focuses on the following features in the above-mentioned conventional misrecognition. (1) It is highly possible that the correct character A exists in the candidate character string. (2) In the opposite case, that is, in the case of inputting the character B that is erroneously determined to be the correct answer for recognition, the character A is unlikely to appear as a candidate character string.

Accordingly, the first aspect of the invention outputs the character having the highest degree of similarity obtained by correlating the characteristic amount extracted from the unknown character pattern with the characteristic amount of each character in the dictionary as the recognition result. In the character recognition device that
A table created by combining each character code that is erroneously recognized as a true character code and a character code that is obtained by erroneous recognition as an error character code, and a correlation calculation performed for each unknown character and each character in the dictionary. A means for creating a candidate character string by extracting a predetermined number of the obtained similarities for each character in descending order, and whether the leading character code of the candidate character string corresponds to one of the erroneous character codes in the table. And a means for determining the true character code corresponding to the false character code when there is a corresponding false character code in the table and determining whether or not the true character code is in the candidate character string. And a means for outputting the true character code as a recognition result when there is a true character code in the candidate character string.

A second invention is characterized in that each character code in the first invention is replaced with a character cluster code.

[0008]

In the first aspect of the invention, the table is created by combining the codes of the characters that are likely to be erroneously recognized as the true character codes and the codes of the characters obtained by the erroneous recognition as the erroneous character codes. A candidate character string is created by extracting a predetermined number in descending order of similarity for each character obtained by the correlation calculation performed for each unknown character and each character in the dictionary. It is determined whether the character code at the beginning of the candidate character string corresponds to one of the error character codes in the table, and if it is determined that there is an error character code, the true character code corresponding to the error character code is It is read and it is determined whether or not the true character code is in the candidate character string. When it is determined that there is a true character code in the candidate character string, the true character code is output as the recognition result.

In the second invention, similarly to the first invention, the character cluster code is used instead of the character code for recognition.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a flow chart showing the operation of the embodiment of the present invention. The operation will be described below based on this flowchart. In the figure, from step1 to step3,
Image input, cutout, and recognition are conventional processes. Here, assuming that a character string as shown in FIG. 2 is input as an image, the character is cut out from the first character and the similarity with the dictionary is obtained in order for the feature, and from this, a predetermined number is selected in descending order of similarity. Is extracted and a candidate character string is created. Conventionally, the first character of this candidate character string is output as it is as a recognition result.

Next, the table is searched to determine whether or not the first character of the candidate character string is written in the character column B of the table (steps 4 and 5). This table was created by writing the true code of the character code or the character cluster code that is easily erroneously recognized in the column A, and writing the character code or the character cluster code obtained by erroneously recognizing the code in the column B. It is a thing. Tables 1 and 2 show examples of the tables. Table 1 shows the case of the character code, and Table 2 shows the case of the character cluster code.

[0012]

[Table 1]

[0013]

[Table 2]

In the embodiment, when the input character string shown in FIG. 2 is recognized in order, the result becomes FIG.
Is mistakenly recognized as "ro". When the table is searched for the character code or the character cluster code of this misrecognized "RO", the code of "RO" can be found in the first column B of Table 1 and Table 2. As a result, it is highly possible that the unknown character recognized as “RO” is mistakenly recognized as the character “U” written in the column A.

Next, if the first character of the candidate character string is written in the character field B of the table, after reading the character code or the character cluster code written in the corresponding character field A, the candidate character string is selected. A search is performed to determine whether or not there is a code in character column A (hereinafter referred to as code A) (st
ep6,7). If code A is in the candidate character string,
Since the position of the code A is after the beginning, the code A is moved to the beginning, the code that was the beginning until then is moved to the end, and the code A is output as the recognition result (step).
8).

In the embodiment, the "u" written in the A column corresponding to the "B" in the B column of Tables 1 and 2 is the second in the candidate character strings as shown in FIG. Therefore, “u” is moved to the beginning and “ro” is moved to the end. At the same time, the code of "u" moved to the head is output as a recognition result. In the above embodiment, the correction process when the recognition result “ro” is incorrect is shown, but the process when the recognition result “ro” is correct will be described below.

Assuming that the character string shown in FIG. 5 is input and recognized as shown in FIG. 6, as a result of the table search, there is a character "RO" in the B column of Table 1 and Table 2, but Since the "u" written in the column A corresponding to the "ro" is not in the candidate character string of the character "ro" shown in FIGS. 7 and 8, the leading "ro" of the candidate character string is not included. Is output as a correct recognition result. As described above, in the embodiment of the present invention, even when a font other than the font prepared as the dictionary is input, the recognition accuracy is improved without newly creating the dictionary of the corresponding font. Can be made.

Further, since the table installed in the embodiment has a small capacity compared with the data amount of the dictionary, it is possible to save memory as compared with the case where a new dictionary is created and added. A given storage capacity enables recognition of more fonts. In addition, in the embodiment, the reason why the character cluster code is used in addition to the character code is that there is a possibility that one character is written in a plurality of tables depending on the font, and the capacity of the table is larger than that in the case of only the character code. Is slightly larger, but the accuracy is further improved. Also, in an actual character recognition device, if the dictionary for only one font is used, only the character codes in Table 1 are used,
If it has a dictionary of plural fonts, the character cluster code in Table 2 is used.

[0019]

As described above, according to the first and second aspects of the invention, once a character having a high degree of similarity is obtained, the table is referenced to correct the error so that the character having a different font from the dictionary The recognition accuracy when input is improved. Further, as a result, it is possible to recognize characters in other fonts without adding a new dictionary.

[Brief description of drawings]

FIG. 1 is a flowchart showing the operation of an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing an example of an input image.

FIG. 3 is an explanatory diagram showing a recognition result of the input image of FIG.

FIG. 4 is an explanatory diagram showing a processing example of a candidate character string.

FIG. 5 is an explanatory diagram showing another example of an input image.

FIG. 6 is an explanatory diagram showing a recognition result of the input image of FIG.

FIG. 7 is an explanatory diagram showing correspondence between input characters and candidate character strings.

FIG. 8 is an explanatory diagram showing correspondence between input characters and candidate character strings.

FIG. 9 is an explanatory diagram showing input characters and dictionary characters in comparison with each other.

FIG. 10 is an explanatory diagram showing correspondence between input characters and candidate character strings.

FIG. 11 is an explanatory diagram showing correspondence between input characters and candidate character strings.

Claims (2)

[Claims] 1. A character recognition device for outputting a character having the highest degree of similarity obtained by performing a correlation calculation of a characteristic amount extracted from an unknown character pattern with a characteristic amount of each character of a dictionary as a recognition result. It is obtained from the table created by combining each character code that is easily recognized as a true character code and the character code that is obtained by misrecognition as an error character code, and the correlation calculation performed for each unknown character and each character in the dictionary. A method of creating a candidate character string by extracting a predetermined number of similarities for each character that is obtained and checking whether the character code at the beginning of the candidate character string corresponds to one of the error character codes in the table. If there is a corresponding erroneous character code in the table and the determining means, the true character code corresponding to the erroneous character code is read, and the true character code is the candidate sentence. A character recognition device comprising: means for determining whether or not the character string is in a string; and means for outputting the true character code as a recognition result when the candidate character string has a true character code. . 2. The character recognition device according to claim 1, wherein
A character recognition device characterized in that each character code is replaced with a character cluster code.