JPS60153574A - Character reading system - Google Patents

Abstract

PURPOSE:To attain accurate character reading even if characters having difference size are contacted each other by determining the number of divisions in accordance with the size when the lump of a black string on a character line is larger than a fixed section, and executing different charactor segmenting methods each other. CONSTITUTION:Characters on a form are converted into binary pattern data by a photoelectric conversion circuit and temporarily stored in a pattern memory 12 from an input terminal 11 to a pattern memory 12. A character segmenting part 13 segments a row pattern including characters for one line by the pattern memory 12, and while moving a remark point in the row direction, executes the scanning of the column direction and takes out data (black string data) obtained by indicating a part including the pattern by the number of black picture elements. In addition, the character segmenting part 13 segments an individual pattern or a forced separating pattern as a discrimination pattern from the row pattern on the basis of the black string data and sends the segmented pattern to a feature extracting part 14. The feature extracting part 14 extracts the feature of the character, a discrimination part 15 collates the extracted feature with a discrimination dictionary part 16 and a character decision part 17 processes the sent data and outputs the selected character as a character reading result.

Description

[Detailed Description of the Invention] (Technical Field) The present invention relates to a character reading method that can read characters in a document with many touching characters in which the character pitch is equal to the character size with high precision and at high speed. .

(Prior Art) The present inventor previously developed a method for character reading in which characters are extracted character by character from a pattern of character lines obtained by full scanning photoelectric conversion of text on a document, and a predetermined constant value on a character line is extracted. Check the number of point sequence clusters that exist within the interval, and if there are - pieces, is it the interval? - Cut out the chisel turns as one character, and if there are multiple combinations of dots, cut out all of the combination patterns that suitably combine the clusters of dots sequentially. Based on the cutting out process to output, the identification result of the cut out pattern, and the information regarding the cutting out, if the pattern is considered to be a single character pattern, the identification result is output as is, and if it is considered to be multiple patterns, the identification result is output as is. A character reading system has been invented which includes a character determination step of outputting a recognition result corresponding to the combination with the longest pattern width - ξ turn from among the recognition results of each of a plurality of combination patterns. This invention is currently under patent application (Japanese Patent Application No. 57-222489) by the present applicant. This prior invention has the advantage of being able to accurately and quickly read documents with uneven character pitches, documents with a mixture of full-width and half-width characters, etc., but when characters of different sizes come into contact or In some cases, the desired character reading result may not be obtained, such as when one of the characters is crossed out.

(Objective of the Invention) In view of the above-mentioned point of view, the object of the present invention is to achieve even higher accuracy and even when characters of different sizes touch each other or when one of the touching characters overlaps. The purpose is to provide all character reading methods that can be read at high speed.

(Structure of the Invention) In order to achieve the above-mentioned object, the first invention is a character reading method that performs character recognition by cutting out each character from a black and white binary character line pattern obtained by full scanning photoelectric conversion of characters on a form. In this method, the number of clusters of dots that exist within a predetermined interval on a character line is counted, and if the cluster of dots on the character line is larger than a predetermined interval, the number of clusters of dots on the character line is counted. A character cutting process in which the number of divisions is fully divided according to the size, and a plurality of different character cutting methods are performed to output information on character cutting and turns in forcedly separated cutouts, and identification of the cutouts and ξ turns. Using the results and the unknown information regarding the extraction, the character extraction method that shows the highest accuracy value among multiple character extraction methods is regarded as the optimal character extraction method, and all identification results obtained with that character extraction method are The second invention is characterized by a character determination step that outputs a character reading result, and the second invention recognizes characters by cutting out each character from the chisel turn of a black and white binary character line obtained by full scanning photoelectric conversion of characters on a form. In the character reading method that performs
Check the number of clusters of dots that exist within a predetermined interval on the character line, and if the cluster of dots on the character line is larger than the predetermined interval, calculate the number of divisions of the cluster of dots. Multiple types are set depending on the size of the point array block, and different types and types are set.
A character cutting process in which several types of character cutting methods are performed to output all information on character cutting and the forcedly separated cut-out patterns, and the identification results of the cut-out chisels and information on the cutting-out? A character determination process in which the character extraction method that has the highest accuracy value among multiple types of character extraction methods is regarded as the optimal character extraction method, and the identification results obtained by that character extraction method are output as all character reading results. and? It is characterized by having.

(Embodiment) The drawing shows an embodiment of the present invention, in which 11 is an input terminal, 12 is a pattern memo 1, 13 is a character cutting section, 14 is a feature extraction section, 15 is an identification section, and 16 is an identification section. 17 is a character determining section; 18 is an output terminal;

The operation of each part in the above configuration will be fully explained below. First, the characters on the form are converted into a binary black and white pattern by a photoelectric conversion device (not shown), and this is sent to the input terminal ll.
It is temporarily stored in the nomiturn memory 12 via the money. The character cutting unit 13 cuts out a line pattern 20 including one line of characters as shown in FIG. 2 from the turn memory 12, and then moves in the direction of all the lines of interest (in the direction of the arrow X in the figure) while Scanning is performed in the column direction (direction of arrow Y in the figure), and the part where the pattern exists is represented by the number of black pixels, and the part where the pattern does not exist is 0.
Data displayed as (hereinafter referred to as point sequence data)
'30 'r Take out. Furthermore, the character cutting unit 131d executes all character cutting processes to be described later based on the black column data 30, and cuts out all of the line patterns 20, individual chisel turns 21, forced separation chisel turns 22, etc. as ξ turns for identification. Information (character cutting position in row/zo turn 20, number N of clusters of point sequences within a certain interval α,
An operation number DNO indicating how many times the operation to detect a sunspot sequence cluster is repeated, an A turn number PNO created by combining the constant interval α constant interval clusters, the number M of cranes for forced separation, and the type of forced separation. It is sequentially sent to the feature extraction unit 14 as a pair of character nomiturn data for identification along with the number of separations L) for each character. The feature extraction unit 14 extracts character features from the received character pattern. Then, the feature data and information regarding character extraction are sent to the identification unit 15. The identification unit 15 performs a comparison with the identification dictionary unit 16, sequentially identifies characters and ξ turns for identification, and combines the identification results (for example, character code and similarity) with information regarding character extraction. The data is sequentially sent to the character determination unit 17 as a pair of data. Character determination section 17
Then, the sent data is subjected to processing to be described later, and the selected data is outputted to the output terminal 18 as a character reading result.

The character cutting process for creating forced separation/ξ turns 22 in the character cutting section 13 is as shown in FIG. In the row ξ turn 20, if there is no cluster of point sequences within the fixed interval α, or if there is one cluster or a plurality of clusters, this will not be discussed since this is detailed in Japanese Patent Application No. 57-222489. If the cluster of point sequences is larger than the fixed interval α, the number of character patterns to be cut out L from the size of the cluster of point sequences
K is determined (for example, when the width of a cluster of dots BLMA≦BL<(L0-!-)MA, L is the number of all characters to be extracted.Also, the relationship with the average character width MA' of half-width characters is (L'-1 ) to M'≦BL<(L'shi-!-) When MA', 2 2 L' is also the number of characters to be cut out), this value? Use this to cut out characters. At this time, there are multiple types of character cutting methods (M type)
(For example, when M = 3, the method of cutting out the average character width MA years from the starting position of the block ■, the method of cutting out the average character width MA years units from the ending position of the block ■, the method of cutting out the average character width MA years units from the starting position of the block There are methods such as (■) that consider the points that divide the space into equal parts as cropping candidate positions, etc.), the number of character patterns to be cut out, the number of types of cutouts M, and the character patterns that are cut out by one cutting method to be sequentially attached. Pattern number PN'0
(PNO=1 to L) 1 to + for each method of cutting out 9 characters
Operation number DNO incremented by 1 (DNO=1
- M) and the like and the cut out chisel turns are sent to the subsequent identification section 15. Identification section 15
Then, the features of the character pattern extracted by the feature extraction unit 14 are compared with the character features prepared in the identification dictionary unit 16, and those with a degree of similarity of more than a certain value are selected as identification results, along with information regarding character extraction. Character code, similarity, etc.? It is output to the character determination section 17. In the character determination section 17, the identification section 1
The entire character determination process shown in FIG. 4 is performed based on the information regarding character extraction sent from 6 and the identification results.

In FIG. 4, it is determined from the information regarding character cutting sent from the identification unit 16 whether or not the identification result is a forced separation nomiturn identification result, and forced separation is performed.

In the case of a turn, the identification results are temporarily stored in the entire buffer,
When the final identification results of consecutive forced separation -ξ turns are sent, a selection process is performed in which all highly accurate character cutting methods are selected from among the notations stored so far and outputted as the reading results.

Next, the process of character extraction and character determination will be explained for the example of row no ξ turf 20 trays shown in FIG.

Among these, the selection process in character determination can be performed using a method that uses similarity all around as the identification result or a method that uses the priority (rank) of the identification result, but here, similarity will be used for explanation. In the row ξ turf 20, the character segmentation unit 13 performs forced separation processing for the o-turn "direction" in the target section 2 because the point sequence data 30 is thicker than the predetermined constant section α. At this time, since the cluster of dots BL is almost equal to 2MA, the number of divisions is L = 2, and if the above example M = 3 is adopted as the character cutting method, the output pattern from the character cutting part 13 is shown in Figure 5. As shown in "Yu J, rt", r
There are six types of chimiturns: ``Hou'', ``Sei'', ``Right'', and ``Sou'', and information on how to cut out each character. The character determination unit 17 completely detects that this section is a forced separation nomiturn section, and selects the one with the highest accuracy among the identification results. select. In other words, what is the average value of the similarity for the identification results of Tano ξ turns with a Nomiturn number assigned to each action number? The method with the highest value of Na (item number 2 in the example in Figure 5) is adopted as the optimal character extraction method.
At that time, the identification results "Ko" and "Tei" are sent to the output terminal 18 as all character reading results.

In this way, according to the above embodiment, it is possible to distinguish whether it is a single character pattern or a plurality of characters that touch each other depending on the size of the block of dots in the character line, so that it is cut out as a - character. It is possible to distinguish whether it is an interval or an interval that should be cut out as multiple characters, and since it is possible to perform multiple character cutting numbers and multiple types of character cutting methods, not only full-width characters but also half-width characters can touch each other. Can it be cut out and character reading accuracy? can be raised. In addition, since the character cutting unit 13 only needs to mechanically cut out ξ turns according to the size of the cluster of dots, it is necessary to cut out the entire character reading process. It can be configured as a no-no-b line configuration, and the processing speed can be increased.

In the character cutting step in the embodiment, a plurality of types of division numbers of a dot sequence block may be set depending on the size of the dot sequence block, and a plurality of different character cutting methods may be performed. In this way, the reading accuracy is further improved.

(Effect of the invention).

As explained above, according to the present invention, character recognition is possible by cutting out each character from a character line pattern obtained by scanning and photoelectrically converting a document on a form. In the character extraction method to be used, all sizes of clusters of dots on a character line are checked, and if the size is larger than a predetermined interval, the number of divisions is determined according to the size of the cluster of dots, and multiple types of different types are determined. How to cut out characters in 7 ways? - Regard it as a character pattern and forcibly cut it out, output the cut-out Tano ξ-turn and information about the cut-out, determine that it is a forced separation chisel-turn from the information about the cut-out of the cut-out chisel-turn, and identify the forced-separation chisel-turn. A highly accurate extraction method is detected from the results, and the identification result is output as a single character reading result. The document contained therein can be read uniquely without performing complicated processing, speeding up the processing, and achieving high accuracy. In addition, in the case where multiple types are set depending on the number of divisions of a block of dots and the size of a block of all black lines, and several different character extraction methods are executed, there are many different forced separation patterns. This has the advantage that the reading accuracy can be further improved.

[Brief explanation of the drawing]

Fig. 1 is an example of a character reading device to which the method of the present invention is applied. FIG. 2 is an explanatory diagram showing a row no ξ turn and its point sequence data, and FIG. 3 is a block diagram showing the character cutting unit 13.
4 is a flowchart of the character determination unit 15, and FIG. 5 is a flowchart of character cutting and identification for rows and o-turns. How is the character determination process executed? FIG. 11...Input terminal, 12...Sξ turn memory, 13
...Character extraction section, 14...Feature extraction section, 15...
-Identification unit, 16...Identification dictionary unit, 17...Character determination unit, 18...Output terminal. Patent applicant Nippon Telegraph and Telephone Public Corporation agent Patent attorney Yoshi 1) Sei, Takashi

Claims (2)

[Claims] (1) Black and white obtained by scanning and photoelectrically converting characters on a form 2
In a character reading method that performs all character recognition by cutting out each character from the chisel turn of a character line, the number of clusters of point sequences existing within a predetermined interval on the character line is checked, and the number of clusters of point sequences on the character line is calculated. If the block is larger than a predetermined certain interval, how many times should the point sequence be divided according to the size of the block? A character cutting step in which a plurality of different character cutting methods are performed to output information on character cutting and a forcedly separated cutout pattern; and a character determination step in which a character extraction method that exhibits the highest accuracy value among the character extraction methods is regarded as the overall optimal character extraction method, and an identification result obtained by that character extraction method is output as a character reading result. A character reading method featuring: (2) Black and white 2 obtained by scanning and photoelectrically converting characters on a form
In a character reading method that performs character recognition by cutting out each character from the number turn of a character line, it is possible to count the number of clusters of dots that exist within a predetermined interval on a character line, and to identify the number of dots on a character line. If the block is larger than a predetermined certain interval, how many times should the block of points be divided? Is it possible to set multiple types depending on the size of the cluster of dots and use multiple different character extraction methods to forcefully separate the information related to character extraction? Using the character cutting process to output, the identification result of the cutting nomimeter, and information regarding the cutting,
A character determination step in which a character extraction method that shows the highest accuracy value among multiple types of character extraction methods is regarded as the optimal character extraction method, and the identification results obtained by that character extraction method are output as all character reading results. A character reading method characterized by: