JPH05174195A - English character recognizing device - Google Patents

Abstract

PURPOSE:To shorten processing time in a processing circuit for a spelling check or the like by effectively ordering candidate character strings in the case of executing the spelling check. CONSTITUTION:This device is provided with a character recognition part 5 to output the plural candidate characters based on the graphical features of a character area, and a spelling check part 7 checks the candidate characters outputted from the character recognition part from the high order of word candidates while storing those characters for one word so as to calculate probability as the word by adding a corrected value calculated based on the product of recognizing accuracy for respective characters in respect to the combination and adjacent frequency for each statistically obtained character position.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an English character recognition device for recognizing English characters.

[0002]

2. Description of the Related Art In recent years, there has been an increasing demand for using a character recognition device as an input device for a computer or the like, and a character recognition device capable of efficiently obtaining a stable recognition result is indispensable for improving the performance of a system such as a computer. Has become. When a plurality of candidate characters are obtained as the recognition result of one input, the conventional recognition device determines some correct characters from among the candidate characters, and therefore, some recognition characters are used depending on the recognition result of the words before and after the character. A character string was generated, and a character string that would be the correct spelling was determined from the character string by using a method such as spell check, and the result was recognized.

[0003]

However, in the above-mentioned conventional character recognition result, when a character string that is a candidate for a word is generated, the candidate character strings are ranked based only on the accuracy of the recognition result. When a plurality of characters in the same word have many candidate characters, the ranking of the candidate character strings is not always appropriate, and the number of processes such as spell check increases, resulting in an increase in processing time. The present invention solves the above-mentioned conventional problems, and an object of the present invention is to provide an English character recognizing device which requires a small number of times of processing such as spell check and a short processing time.

[0004]

In order to achieve the above object, the present invention provides an English character recognizing device that statistically obtains the product of the recognition accuracy of each character for each combination of word candidate character strings. Processing such as spell check is performed by adding a correction value calculated based on the adjacency frequency for each character position to obtain the accuracy as a word and effectively ranking the candidate character strings when performing spell check. The number of times is reduced to shorten the processing time.

[0005]

Therefore, according to the present invention, the order of the candidate character strings when performing the spell check is ranked based on the recognition accuracy of each character and the adjacency frequency for each position of the characters statistically obtained. An appropriate ranking can be performed, the number of times of processing such as spell check can be reduced, and the processing time can be shortened.

[0006]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows the configuration of an English character recognition apparatus according to an embodiment of the present invention. In FIG. 1, 1 is an image input unit for inputting a recognition target document as a document image, 2 is a sentence region cutout unit for finding a set of character strings from the input document image and outputting a sentence region, 3 is a word from the sentence region A word cutout unit that finds a unit break and outputs a range of one word as a word region. 4 is a character cutout unit that finds a unit break of a character from the word region and outputs a range of one character as a character region. A character recognition unit that outputs a plurality of corresponding candidate characters based on the graphic features extracted from the character recognition unit, 6 stores one word of the candidate characters output from the character recognition unit, and a character string obtained by the combination , The rank as a candidate, by adding the correction value calculated based on the product of the recognition accuracy of each character and the adjacency frequency for each position of the character statistically calculated Word candidate generation unit for generating a word candidates attached, 7 checks the spelling from highest rank among the word candidates, spell checking unit that outputs a correct character string as a recognition result, 8
Is a character adjacency frequency list that stores the adjacency frequency for each character position, 9 is an internal bus that connects the parts 1 to 7, and 10 is an internal bus that connects the word candidate generation part and the character adjacency frequency list. FIG. 2 is an overall flow chart of the character recognition processing.

Next, the operation of the above embodiment will be described with reference to FIGS. The text to be recognized is input as a text image in the image input unit 1 (S11). The input text image is sent to the text area cutout unit 2, the vertical and horizontal black pixel histograms of the text image are obtained in the text area cutout unit 2, and the text area is found based on the histogram. Storing position information as internal data (S1
2). Next, the position information of the text area is sent to the word area cutout unit 3, and the word area cutout processing is performed on the inside of the text area. The word area cutout unit 3 pays attention to the fact that the space before and after the word is larger than the space between the characters in the word, and cuts out a character string sandwiched by spaces of a certain width or more as a word area. The position information of all word areas in the text area found by the text area cutout unit 2 is obtained and stored as internal data (S13). Next, the position information of the word area is sent to the character area cutout unit 4, and the character area cutout process is performed on the word area. The character area cutout unit 4 pays attention to the change in the histogram of black pixels in the word area, separates the string of characters in the word area into character units by cutting out a part where the histogram is less than a certain value as a character-to-character separation, and cuts out as a character area. .. The position information of all character areas in the word area found by the word area cutout unit 3 is obtained and stored as internal data (S14). The character recognition unit 5 performs recognition processing on a character-by-character basis based on the position data of the text area, word area, and character area obtained in S12 to S14, and the recognition candidate character that is the result of the recognition processing. Recognition accuracy as internal data 1
The word is stored (S15 to S17). Hereinafter, it is assumed that the recognition candidate character and the recognition accuracy for one word are obtained as shown in Table 1.

[0008]

[Table 1]

Word candidates are obtained by combining the obtained recognition candidate characters (S18). In this example, the first, second, and third characters have one candidate character, but the fourth and fifth characters have two characters, so the following four combinations can be considered as word candidates.

[0010]

[Outer 1]

Next, the accuracy as a word, that is, the word accuracy is calculated for all the obtained word candidates (S18-
S21). The word accuracy is obtained by summing the product of the recognition accuracy of each character forming the word and the correction value calculated based on the adjacency frequency for each position of the character, and the recognition accuracy is obtained by the character recognition process. The number of characters that are adjacent to each other and the position of each character are n, including the spaces at both ends of the English word, as shown in Table 2 below.
When the character is the nth character, the adjacency frequency of the characters is statistically calculated for all the characters between the first and second characters, the second and third characters, ... (n-1) th and nth characters. It is found by fitting the obtained (n-1) adjacent frequency data between each character. In addition,
The character adjacency frequency data is stored in the character frequency list.

[0012]

[Table 2]

First, the product of the recognition accuracy of each word candidate is obtained.
(S19). The closer the recognition accuracy is to 1, the higher the probability that the recognition result is correct, and the recognition accuracy of the entire word is the product of all the constituent characters. The results obtained are shown below. Word candidate 1: 1.0 × 1.0 × 1.0 × 0.7 × 0.6 = 0.42 Word candidate 2: 1.0 × 1.0 × 1.0 × 0.7 × 0.4 = 0.28 Word candidate 3: 1.0 × 1.0 × 1.0 × 0.3 × 0.6 = 0.18 Word candidate 4: 1.0 × 1.0 × 1.0 × 0.3 × 0.4 = 0.12 Next, a correction value based on the adjacent frequency is obtained (S20). Now, the adjacency frequency between characters is determined from the adjacency frequency data determined by the position of the target two characters (what character it is), and the character with multiple candidates (in this example, the (5th character and 6th character) The product of the adjacency frequency with the preceding character and the adjacency frequency with the succeeding character is the character adjacency value,
If the word adjacency value is the product of the adjacency values of all characters that have multiple candidates in the word candidate, then the correction value is the word adjacency value of each word candidate divided by the sum of the word adjacency values of all word candidates. It becomes a thing. Below, a part of the adjacency frequency data required in this example is shown, and the result of obtaining the word adjacency value of each word candidate is shown. (1) Adjacent frequency data:

[0014]

[Outside 2]

(Iii) 6th to 7th character spacing c, e, (blank) adjacency frequency ・ c and blank adjacency frequency is 7 ・ e and blank adjacency frequency is 446 (2) Word candidate 1 (appic): (i) Character adjacency value of the fifth character The adjacency frequency of p and i is 8, and the adjacency frequency of i and c is 55, so the character adjacency value is 440. (ii) Character adjacency value of the 6th character The adjacency frequency of i and c is 55, and the adjacency frequency of c and blank is 7
Therefore, the character adjacency value is 385. (iii) Word Adjacent Value The word adjacent value is the product of the character adjacent value of the fifth character and the character adjacent value of the sixth character, that is, 169400. (3) Word candidate 2 (appie): (i) Character adjacency value of the fifth character Since the adjacency frequency of p and i is 8 and the adjacency frequency of i and e is 37, therefore the character adjacency value is It is 296. (ii) Character adjacency value of the sixth character The adjacency frequency of i and e is 37, and the adjacency frequency of e and space is 446. Therefore, the character adjacent value is 16502. (iii) Word adjacency value The product of the character adjacency value of the fifth character and the character adjacency value of the sixth character, that is, 4884592.

[0016]

[Outside 3]

(Iii) Word adjacency value The product of the character adjacency value of the fifth character and the character adjacency value of the sixth character, that is, 12992.

[0018]

[Outside 4]

(Iii) Word adjacency value The product of the character adjacency value of the fifth character and the character adjacency value of the sixth character, that is, 67049856. Next, the results of corrections based on the word adjacency values are shown. The correction value is obtained by dividing the adjacent value of each word candidate by the sum of the adjacent values of each word candidate. Since the sum of adjacent values of each word candidate is “72116840”, the correction value of each word candidate is: correction value of word candidate 1 = 169400/72116840 = 0.0023 correction value of word candidate 2 = 4884592/72116840 = 0.0677 word candidate Correction value of 3 = 12992/72116840 = 0.0002 Correction value of word candidate 4 = 67049856/72116840 = 0.9297. Next, the word accuracy of each word candidate is calculated. Since the word accuracy is the recognition accuracy plus a correction value, the word accuracy of word candidate 1 is 0.42 + 0.0023 = 0.4223, and the word accuracy of word candidate 2 is 0.28 + 0.0677 = 0.3477. Is 0.18 + 0.0002 = 0.1802, and the word probability of the word candidate 4 is 0.12 + 0.9297 = 1.0497. Next, the spell check unit 7 performs spell check processing in order from the word with the highest word probability of each word candidate obtained by the above processing. The spell check process determines whether or not the character string of a word candidate is correct as an English word. If the check result is correct, the word candidate is output as a recognition result, and if it is incorrect, the next word candidate is checked. .. If all the word candidates are incorrect, the one with the highest degree of recognition among the word candidates is output as the recognition result (S22 to S26). Therefore, the order of spell checking is from the word with the highest word accuracy,

[0020]

[Outside 5]

The character recognition processing of the given sentence image is performed by repeatedly performing S15 to S26 described above for all word regions.

[0022]

As is apparent from the above embodiment, the present invention determines the rank of candidate character strings when performing spell check, the recognition accuracy of each character, and the adjacency frequency for each character position statistically obtained. Therefore, the number of processes such as spell check can be reduced and the processing time can be shortened.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an English character recognition device according to an embodiment of the present invention.

FIG. 2 is a flowchart of the entire character recognition process in the embodiment of the present invention.

[Explanation of symbols]

1 ... Image input section, 2 ... Document area cutout section, 3 ... Word area cutout section, 4 ... Character area cutout section, 5 ...
Character recognition unit, 6 ... Word candidate generation unit, 7 ... Spell check unit, 8 ... Character adjacency frequency list, 9, 10 ... Internal bus.

Claims (1)

[Claims] 1. An image input unit for inputting a document to be recognized,
A text area cutout unit that outputs a text area from the input document image, a word area cutout unit that outputs a word area from the text area, a character area cutout unit that outputs a character area from the word area, and a graphic feature of the character area A character recognition unit that outputs a plurality of corresponding candidate characters based on the above, and one word of the candidate characters output from the character recognition unit are accumulated, and the recognition angle of each character is recognized with respect to the character string obtained by the combination. By adding the correction value calculated based on the adjacency of each character position and the statistically calculated adjacency frequency for each character position, the word candidate with the rank as a candidate is generated by obtaining the accuracy as a word. An English character recognition device comprising a word candidate generation unit and a spell check unit that checks spelling from the word candidate having the highest rank and outputs a correct character string as a recognition result.