JPH0373086A - Character recognition post-processing system - Google Patents

Abstract

PURPOSE:To perform the correction processing of a character with high accuracy by performing the selection of the optimum word candidate out of the characters with the number of strokes near to that of the character with a prescribed processing even when it includes the character not being recognized precisely. CONSTITUTION:The subject system is provided with a word dictionary 2, a word dictionary collation part 3 which collates the word dictionary 2 with a character recognition result 1, a stroke number table 4 on which the numbers of strokes of all the character categories are recorded, and a stroke number collation part 5 operated under a prescribed condition by receiving a word outputted from the word dictionary collation part 3. The character recognized at the word dictionary collation part 3 is collated with the word dictionary 2, and one or plural words with the highest priority are outputted. When plural words outputted from the stroke number table 4 on which the numbers of strokes of all the character categories and the word dictionary collation means 3 exist and no character exists in the plural words in a candidate character set that is the recognition result at the same part, the word with the nearest mean value of the number of strokes can be detected precisely at the stroke number collation part 5. In such a manner, the character with the number of strokes ear to that of the character not being recognized precisely can be selected, which improves correction accuracy.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention utilizes the number of strokes of a candidate character as information in a character recognition device, etc., when the number of strokes of a candidate character as a result of character recognition is included. The present invention also relates to a character recognition post-processing method that selects and corrects accurate candidate characters with a similar number of strokes to improve KR recognition accuracy.

[Prior Art] Generally, a character recognition device outputs a plurality of candidate characters and their distance values or similarities for one input character as a recognition result. In conventional character recognition post-processing, this recognition result is compared with a word dictionary, and if each character in the word dictionary exists in the candidate character set at the corresponding position, that word is determined to be correct. However, this method has the drawback that accurate character recognition output cannot be obtained if the correct character is not included in the candidate characters of the recognition result.

[Problem to be Solved by the Invention] The present invention solves the problem that, in a character recognition post-processing method, the number of strokes of an input character of a character recognition device and the number of strokes of a candidate character as a recognition result often do not change much even if they are recognized incorrectly. The purpose of this paper is to improve correction accuracy by selecting characters with a similar number of strokes even when characters that cannot be accurately recognized are included.

Means for Solving the Problems] In order to solve the above problems, the present invention provides a word dictionary, a word dictionary matching unit that matches the word dictionary and character recognition results, a stroke number table that records the number of strokes of all character categories, It includes a stroke number matching section that receives the word output from the word dictionary matching section and operates under predetermined conditions.

[Function] In the word dictionary matching section, the recognized characters are compared with the word dictionary, a priority is added to each word, the word or words with the highest priority are output, and the number of strokes for all character categories is recorded. If there are multiple words output from the table and the word dictionary matching means, and the corresponding character does not exist in the candidate character set of the recognition result at the same location, the stroke number matching unit uses the stroke number table to The number of strokes of the characters in the set is compared with the average value of the number of strokes of the single or plural candidate characters at the top of the position in the recognition result, and the word having the closest average value at the position is output as the correct answer.

[Example] An example of the present invention will be described below with reference to the drawings.

FIG. 1 shows an embodiment of a device constituting a character recognition post-processing system targeted by the present invention.

In FIG. 1, 1 is a recognition result, 2 is a word dictionary, 3 is a word dictionary matching section, 4 is a stroke number table, 5 is a stroke number matching section, 6
is the correction result.

The recognition result l from character recognition is first passed to the word dictionary collation unit 3.
is input. FIG. 2 is a flowchart of the processing in the word dictionary matching section 3.

Further, FIG. 3 is a diagram illustrating an embodiment in which words are processed in the word dictionary matching section according to this flow.

In FIG. 2, step S1 recognizes 8B character strings in a character pattern recognition device, cuts them into words in step S2, ranks recognition candidates in S3, calculates the distance between each character, and stores them in a predetermined memory as recognition results. Store in place. S4 is a step of comparing the recognition result 1 with words stored in the word dictionary 2, and S5 is a step of calculating the number of matched characters and total distance for each matched word and storing it in the intermediate result storage section of the word dictionary matching section. S6 is a step of sorting the data processed in S5 in descending order of the number of matching characters. After extracting the group with the largest number of matching characters in S7, S8 is a step of sorting the data processed in S5 in descending order of the number of matching characters. Sort them in order and store them in a memory location. The data processed in this way determines the number of words with the minimum total distance value in S9, and if it is singular, generates a judgment output that sends only that word to the next stage stroke number matching unit,
If the judgment output is ■, that is, if there are a plurality of words, then in addition to those words, the character string extracted from the first candidate character of the recognition result is sent to the stroke number matching unit.

In FIG. 3 illustrating an example of processing, 11 is an example of the recognition result sent from the character recognition device, and the correct answer is "Saiwai-cho, Kawaguchi City." In example 11, the correct answer is not included in the candidate character set resulting from recognition of "mouth".

21 is an example of internal representation of the word dictionary 2. The word dictionary collation unit 3 collates the recognition result 1 with each word in the word dictionary 2, temporarily stores the processing results in the intermediate result storage unit 31, and then sorts them in descending order of the number of matched characters. After extracting large groups, the groups are sorted in descending order of total distance value and word dictionary matching result section 3
Store in 2. If there is a single word with the smallest total distance value, only that word is sent to the stroke number matching section 5, which is the next stage of processing. If a plurality of words exist, a character string made from all of the plurality of words and candidate characters of a predetermined rank as a result of character recognition is passed to the stroke number matching unit 5. In this example, the word candidates are rJl Saki City Saiwai-cho”, “Kawaguchi City Saiwai-cho”,
"Kawagoe City Saiwai-cho" and the candidate characters as a result of recognition are sent to the stroke number matching section 5.

FIG. 4 is a flowchart of the processing in the stroke number matching section 5, showing data such as words to be stored in each register and the contents of the processing at each stage.

FIG. 5 shows an example of processing according to the processing flow of FIG. 4, in which the stroke number matching unit 5 registers three word candidates sent from the word dictionary matching unit 3 in a two-dimensional array. The candidate characters resulting from the recognition are also stored in the first to third rows of the two-dimensional array register b-1. Next, compare the contents of register me 1 vertically from the beginning, from the first character of the first row to the first character of the second row, and if all the characters in each column are the same, then ○, otherwise Record the x symbol in that column. In the example in Figure 5,
Since the characters present in the second column from the beginning are different, an x symbol is recorded.

After that, in register me 1, write the characters in the column in which the cross symbol is recorded in register me 2, and write the characters in the column in which the cross mark in register me 1 is recorded in the characters in register t)-1 Candidate characters belonging to the same column number are written into register b-2 in order from the top, ie, in the order of the first row, the second row, and so on.

Although the number of characters to be written into the register b-2 can be set freely, in this embodiment, it is set to write the top three candidate characters. In the example of FIG. 5, "Kuchi", "Koshi", and "Saki" are written in register me2, and "ha", "ha", and "ni" are written in register b-2.

Next, the number of strokes of each character of register me 2 is checked with reference to the table 41. Then, the number of strokes of each character in register b-2 is checked with reference to the number of strokes table 41, and each number of strokes is calculated by the divider 55.
, and divide the sum of the number of strokes by the sum of the number of characters to find the average value. Next, compare that average number of strokes with the number of strokes of each character in Register Me 2, select the character in Register Me 2 that has the closest number of strokes to the average number of strokes of the characters in Register b-2, and select the word candidate containing it as the correct word. Output as . In the example shown in FIG.
"Kawaguchi City Saiwai-cho" is output as correction result 6.

In the embodiment, the suitability of candidate characters is determined using distance values, but similar processing can be performed using similarity.

The device for implementing the present invention can be realized as an integrated circuit that incorporates a dedicated circuit and can input the character 1 EWi result and output the correct word.
Further, it is also possible to add a control circuit with a function capable of achieving the object of the present invention to a general-purpose information processing circuit, and store necessary character information and the like in a storage circuit.

[Effects of the Invention] As described above, when performing character recognition post-processing, even when characters that are not accurately recognized are included, the present invention can select the optimal one from among characters with a similar number of strokes through predetermined processing. It is possible to select word candidates, and thereby, after character recognition, correction processing for characters that cannot be accurately recognized can be performed with high accuracy.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an example of a character recognition post-processing method targeted by the present invention, FIG. 2 is a diagram for explaining an example of processing in the word dictionary matching section 3 of FIG. 1, and FIG. is the first
FIG. 2 is a diagram showing the contents of each internal register and a stroke number table, and the processing, in order to explain an example of processing in the stroke number matching unit 5 shown in the figure. 1: Recognition result 2: Word dictionary 3: Word dictionary collation unit 4: Stroke count table 5: Stroke count collation unit 6: Correction result 11; Example of recognition result sent from character recognition device 21: Internal representation of word dictionary 2 Example: Intermediate result storage section: Word dictionary matching result section Two-sided number table: Register a-1: Register b-1 Register me 2: Register b-2: Divider

Claims (1)

[Claims] A character recognition device that recognizes an input character pattern and generates a distance value or similarity between a single or multiple candidate characters and a standard pattern, compares the recognition results with a word dictionary and determines priorities. There is a word dictionary matching unit that outputs one or more words with the highest priority added to each word, a stroke count table that records the number of strokes of all character categories, and multiple words output from the word dictionary matching unit.
In addition, if the plural words are in the same location and the corresponding character does not exist in the candidate character set of the recognition result, the number of strokes of the character in the word set and the number of strokes of the single or multiple candidate characters at the top of the recognition result at the location are determined. A character recognition post-processing method comprising: a stroke count comparison unit that compares the average value and outputs the word having the closest average value at the location as the correct answer.