JP2891368B2 - Post-processing method of character recognition result - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to character recognition, and more particularly to post-processing for correcting character recognition results.

[Conventional technology]

Since it is extremely difficult to completely eliminate erroneous recognition by character recognition on a character-by-character basis, in Kanji OCR,
Error correction (post-processing) of character recognition results is often performed by word matching, morphological analysis, or the like.

Conventionally, a post-processing method for a general document is such that after recognition of an area to be recognized (eg, a paragraph) is completed, post-processing is performed on all character strings in the area, assuming that the area has no line break. It is supposed to do. Alternatively, there is a method in which a sentence is used as a post-processing unit. In this method, a character string from the beginning of a sentence to a punctuation mark or a reading point is read and processed after ignoring line feed.

Examples of publicly known materials relating to post-processing of character recognition results of Japanese documents by word matching and morphological analysis include, for example, “Nishino et al .:“ Implementing Post-processing of Japanese Document Readers ”, Natural Language Processing 64-6 (November 20, 1987) ), Pp. 45-52 ".

[Problems to be solved by the invention]

However, according to the post-processing method in which a paragraph or a sentence is a processing unit, when a paragraph or a sentence is long, the start of post-processing is waited until the recognition result is completed, so that the processing efficiency is poor. Requires a large memory.

Therefore, an object of the present invention is to provide a post-processing method for character recognition results that can solve such a problem.

[Means for solving the problem]

The present invention is based on a process of reading a character recognition result line by line and correcting an error by word matching or morphological analysis. The recognition result character string of one line to be processed is divided into a plurality of parts of a word or a phrase. From the delimiter part up to the last one as the analysis range, word matching and morphological analysis are performed on the character string at each delimiter part of the analysis range to correct errors, and the character string at the last delimiter part is Is moved to the beginning of the recognition result character string, and is newly set as a processing target of the next line.

In addition, the present invention is also based on a process of reading a character recognition result line by line and correcting an error by word matching or morphological analysis.
Correct the error by performing word matching and morphological analysis in order from the beginning, move the characters or character strings at the end of the line remaining unanalysable to the beginning of the recognition result character string on the next line, and set It is characterized by doing.

(Operation)

One line of a character string is analyzed by being divided into words, phrases, and the like. One word may be split at the end of a line and at the beginning of the next line. A word broken by such a line break cannot be correctly analyzed if it is not taken into account, and if it is misrecognized, the correction fails.

However, according to the present invention, although post-processing is performed on a line-by-line basis, a word or phrase delimiter at the end of a line or an unparseable part at the end of a line is moved to the beginning of the next line, and post-processing of the next line is performed. Therefore, even for a word split by a line feed, the erroneous recognition can be correctly corrected.

In general, in a character recognition device, “area extraction or area specification” → “line extraction” → “character extraction” →
Processing proceeds in the order of “character recognition” → “post-processing”. Therefore, if the unit of the post-processing is one line, the post-processing of a certain line and the character recognition processing of the next line are executed in parallel, thereby reducing the waiting time of the post-processing and speeding up the entire processing. can do. Further, since the number of characters to be printed or written on one line is almost fixed, the amount of memory required for storing the recognition result can be reduced.

〔Example〕

 Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 shows a block diagram of the kanji OCR according to the present invention.
The overall configuration and processing flow of this kanji OCR is the same as in the past. That is, the image input unit 1 reads a target image from the scanner (or another input device) 2 and stores the target image in the image memory 3. The range designating unit 4 automatically designates a range to be recognized from the image input to the image memory 3, or designates the range by designating means. The cutout unit 5 cuts out a line image from the area whose range is designated, and cuts out an image of each character from the line image. Using the dictionary stored in the character dictionary memory 7, the recognizing unit 6 performs character recognition on each of the extracted character images, and stores the recognition result in the recognition result memory 8. The post-processing unit 9 reads the recognition result character string from the recognition result memory 7, analyzes the character string by referring to the word dictionary and the grammar dictionary stored in the word dictionary / grammar dictionary memory 10, and corrects an inappropriate portion as a sentence. I do. The post-processing unit 9 is a part directly related to the present invention, and details thereof will be described later for each embodiment. The output unit 11 outputs the result of the post-processing to the CRT or the output file 12.

Embodiment 1 FIG. 2 shows a block diagram of the post-processing unit 9 in this embodiment. When the character recognition for one line is completed, the recognition result reading unit 21 reads the recognition result for one line from the recognition result memory 8 and stores it in the line recognition result memory 22. If the end analysis part at the end of the previous line has been transferred to the head of the line recognition result memory 22, the writing is performed from the next character position.

Next, the provisional phrase segmentation unit 23 converts a one-line recognition result character string in the line recognition result memory 22 into a plurality of analysis units as an analysis unit according to a position at which a character type changes (for example, switching from hiragana to non-hiragana). Divide into parts. The part separated in this way is called a “provisional clause”.

The analysis range determination unit 24 sets the analysis target range from the first character position of the line recognition result memory 22 to the character position before the last provisional phrase in consideration of the position of the provisional phrase.

The sentence analysis / correction unit 25 uses the word dictionary and the grammar dictionary in the word dictionary / grammar dictionary memory 10 for each provisional phrase in the set analysis target range, and performs word matching and morphological analysis to correct erroneous recognition. Is performed, and the recognition result in the line recognition result memory 22 is correctly rewritten. The output unit 11 outputs the result corrected in this way.

When the analysis and correction are completed up to the last provisional clause in the analysis target range, the final clause writing unit 26 reads the recognition result of the provisional clause at the end of the line in the line recognition result memory 22 (the part excluded from the analysis and correction target), Perform the operation of moving to the beginning. That is, the provisional clause at the end of the line may be separated by a line feed, but this is a provisional clause at the beginning of the next line and is subjected to post-processing of the next line.

In addition, during the execution of such post-processing for one line,
The character recognition processing for the next and subsequent lines is executed in parallel.

Here, the contents of the post-processing will be described using the following target text as an example.

First, it is assumed that the following recognition result is obtained for the first line.

“Character recognition technology is a modern information processing place” This is divided into temporary phrases as follows by character type change (Hiragana → Non-Hiragana).

 “Character recognition technology / modern / information processing” And the analysis target range is as follows.

"Character recognition technology / modern" Now, the word "information processing (physical)" that was split by line breaks
Is excluded from the analysis target in the first row.

When the analysis and correction of the analysis target range on the first line is completed and the result is output, the unprocessed provisional clause at the end of the line is transferred to the head of the line recognition result memory 22, and the character positions thereafter are shifted. Is cleared.

The process moves to the second line, and the result of the recognition is stored in the line recognition memory.
Since it is written continuously from the part moved from the previous line of 22,
The recognition result on the second line is as follows.

"It is a flower-shaped technique of information processing." Thus, the word "information processing", which has been split over two lines, is united and can be analyzed and corrected.

Then, in the same way, break into temporary clauses, set the target range,
Perform analysis and correction. However, since the last character of this line is a punctuation mark, it is determined that word division has not occurred at the end of the line, and the last provisional clause of the line is also set as the analysis target range. The same applies if the last character on the line is a reading point.

Embodiment 2 FIG. 3 shows a block diagram of the post-processing unit 9 in this embodiment. When the character recognition for one line is completed, the recognition result reading unit 31 reads the recognition result for one line from the recognition result memory 8 and stores it in the line recognition result memory 32. If the unanalyzable part at the end of the previous line has been transferred to the beginning of the line recognition result memory 32, the writing is started from the next character position.

Next, the sentence analysis / correction unit 33 uses the word dictionary and the grammar dictionary in the word dictionary / grammar dictionary memory 10 for the contents of the row recognition result memory 32, performs word matching and morphological analysis, and performs row recognition. The recognition result in the result memory 32 is correctly rewritten. The result corrected in this way is output unit 11
Is output by However, if there is a part that cannot be analyzed (a character string in which the corresponding word cannot be found) at the end of the line, the end of the line is marked with “next line”.

The next-line partial writing unit 34 stores a line-recognition result memory.
Move the end of the line marked "Next line" to the beginning of 32, and clear the character positions thereafter. The unparseable part of the end of the current line is now
It is moved to the beginning of the next line and is subjected to post-processing of the next line.

During the post-processing of one line,
The character recognition processing for the next and subsequent lines is executed in parallel.

Here, the contents of the post-processing will be described using the following target text as an example.

First, it is assumed that the following recognition result is obtained for the first line.

“Character recognition technology is a modern information processing” When the character string is analyzed from the beginning by word collation or morphological analysis, the following is obtained (/ is a word break).

"Character / recognition / technology / wa / modern / no / information / process" Since this last character "process" has no corresponding word and cannot be analyzed, this "process" becomes the next line carrying part, and the correction result The output looks like this:

"Character recognition technology is modern information." Next, since "process" is moved to the head of the next line, the recognition result of the second line becomes "processing flower-shaped technology." The word that was connected is connected. Since the last character of this line is a punctuation mark, there is no unparsable part until the end of the line, and processing is performed to the end.

〔The invention's effect〕

As described above, according to the present invention, it is possible to correct the recognition result including a word split by a line feed, and to improve the efficiency of the entire process by executing post-processing and recognition processing in parallel on a line-by-line basis. Is possible, and the unit of post-processing is 1
Since it is a row, the capacity of the storage memory for the recognition result for post-processing can be reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram of a kanji OCR according to the present invention, FIG. 2 is a block diagram of a post-processing unit according to an embodiment of the present invention, and FIG.
FIG. 13 is a block diagram of a post-processing unit according to another embodiment of the present invention. 6: recognition unit, 8: recognition result memory, 9: post-processing unit, 10: word dictionary / grammar dictionary memory, 21: recognition result reading unit, 22: line recognition result memory, 23: temporary Phrase segmentation unit, 24: analysis range determination unit, 25: sentence analysis / correction unit, 26: final phrase writing unit, 31: recognition result reading unit, 32: line recognition result memory, 33: sentence analysis・ Correction part, 34 ... The next line carry part writing part.

Claims (2)

(57) [Claims] 1. A post-processing method for analyzing a character recognition result one line at a time and correcting an error, wherein a recognition result character string of one line to be processed is divided into a plurality of parts of words or phrases, The analysis part is the delimiter part up to the one before, and the word string and morphological analysis are performed on the character string of each delimiter part of the analysis range to correct errors. A post-processing method for a recognition result, wherein the recognition result is moved to the head of the result and is newly processed in the next line. 2. A post-processing method for analyzing a character recognition result one line at a time and correcting an error. Character recognition result post-processing method, wherein the character or character string at the end of the line remaining unanalyzable is moved to the beginning of the recognition result character string on the next line, and is newly processed by the next line. .