JPS59112367A - Reading method of character - Google Patents

Abstract

PURPOSE:To improve reading accuracy and to read out speedily characters by checking the number of lumps of black strings existing in a fixed section on a character line, and when the plural lumps are detected, regarding plural combined patterns obtained by combining the lumps of the black strings successively and properly as one character pattern respectively and segmenting each pattern. CONSTITUTION:Characters on a business form are photoelectrically converted and the converted pattern data are temporally stored in a pattern memory 12 through an input terminal 11. A character segmenting part 13 segments an individual pattern from a line pattern 20 on the basis of black string data 30 temporally stored in the memory 12 and sends a pair of individual data consisting of the individual pattern 21 and information related to the segmenting to a discriminating part 14 successively. After processing the data, a character deciding part 15 outputs only the discriminated results of correct characters to an output terminal 16 successively.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a character reading method that can read characters in fixed documents with high precision and high speed.

In a conventional character reading device, as shown in FIG. The character cutting section 3 cuts out each character from the beginning of the character string at predetermined intervals from the character line,
The character number turn is transferred to the identification section 4 to determine what character it is, and the identification result is outputted from the output terminal 5. For this reason, there was a drawback that unless the character pitch of the form was constant, it could not be read at all.

In addition, the character cutting g++:iK auxiliary identification section 6 of the character reading device is provided, and the auxiliary identification section 6 judges whether only a narrow pattern is one character or a part of a character. If it is determined to be a pattern, the narrow pattern is regarded as one character and character extraction is performed,
Also, if it is determined to be part of a character, the character is cut out in addition to the next chisel turn, so that the character pitch remains constant.
There have already been proposals to make text more readable than on forms.

However, in the above device, the narrow chimi-turn forms a pattern similar to other character patterns (for example, in a horizontally written form, the kanji ``ka'', whose partial ``i'' forms the same character shape as the katakana ``i'', ), when one character is divided into many two turns due to a break in the character, and when characters 2 or more -L written consecutively are judged as one character, it is difficult to obtain sufficient cutting accuracy. It was difficult. When the auxiliary identification section 6 is operated, the operation of the character extraction section 3 is stopped until the determination result is output, which has the disadvantage of slowing down the reading speed. In addition, this reduction in reading speed becomes more noticeable as the number of chisel turns that must be dealt with by the auxiliary identification section 6 increases.

In order to eliminate the above-mentioned conventional drawbacks, the present invention, when a plurality of chisel turns exist within a predetermined interval in a character line on a form, cuts out the characters by sequentially combining the nocturnes and cuts out all of them. It is a system that identifies characters and selects those that are more likely to be characters based on the results.The purpose is to identify characters that can be read accurately and quickly in documents such as forms that do not have a constant character pitch. The objective is to provide a reading method. The drawings will be explained in detail below.

3 to 8 show an embodiment of the present invention, in which 11 is an input terminal, 12 is a chime-turn memory l1, 13
1 is a character cutting section, 14 is an identification section, 15 is a character determination section, 1
6 is an output terminal.

To operate this, the characters on the unshuffled form are converted into pattern data by a photoelectric conversion device (not shown), and this data is temporarily stored in the Q-turn memory 12 via the input terminal 11. The character cutout 1-7 section 13 is the chisel turn memory 12.
Cut out a row pattern 20 containing one line of characters as shown in FIG. The data shows the part where the second turn exists in black ("1" for the 2nd Brigadier's rank), and the part where it does not exist in white ("0" for the 2nd Brigadier's rank).
Hereinafter, this will be referred to as point sequence data. ) Take out 30. Furthermore, the character cutting unit 12 executes a process described later based on the point sequence data 30 to cut out individual chisel turns 21 from the line-cut turns 20, and identifies each individual pattern 21 and information regarding its cutting as a pair of individual data. Section] 4 is sent sequentially. The identification unit 14 sequentially identifies the characters of only the individual nomiturns 21 of the individual data, and sequentially sends the identification result (for example, the character ``co\'') and the information regarding the cutout to the character determining unit 15 as a pair of data. The character determination unit 15 performs processing to be described later on the data and outputs only the identification results of correct characters from the output terminal I6.

[Character Cutting Section] To cut out the individual pattern 21 in step 3, use the beginning of the point sequence data 30 as a starting point (hereinafter referred to as the reference position) and a set of black parts existing within a certain interval α preset in step 17. (Hereinafter, this is referred to as a cluster of point sequences.) If there are -, the section is considered as an individual character 7, and if there are multiple clusters, consecutive clusters of point sequences are - A plurality of tscutans combined by increasing number of -C are each regarded as an individual tscutane of one character, and the position excluding the first cluster of the plurality of point sequences is the standard of the next fixed interval α. It's like a position.

Next, a detailed explanation will be given according to the flow chart shown in FIG. is an individual turn (hereinafter referred to as a combination) ξ turn by a combination of clusters of point sequences. ) is a turn number indicating the order of turns when creating a turn, and the operation number DNO and .xi. turn number PNO constitute information regarding cutting i-.

Count the number N of clusters of point sequences that exist within a certain interval α from the beginning of the uncut point sequence data 30, that is, the starting position of character extraction, and calculate 1, and if N=G, if the interval is a space, DN
O=1, PNO=] and sends the space cut turn to the identification unit 14, and if the section is all a cluster of point sequences, detect the end position of the next cluster of point sequences L7, and select the section. is regarded as a contact character [, and forced separation is performed, and DNO=1. , and sends it to the identification unit 14 with PNO=1 assigned thereto. In addition, when N= ], the section is considered to be an individual character pattern, and the individual pattern VCT) NO-2, PNO=1 is added to the identification part J.
Send to 4.

Furthermore, when [N)], sequentially combine the clusters of dots that appear from the beginning without changing the appearance order of the clusters of black rows, create N silk-combining turns, and use the operation number DNO and 1 to N.
9. For example, when N=3, a cluster of point sequences a, l),
C, in the first process, DNO = 1 and PNO = 1, o-turn "a", DNO = 1 and PNO = 2, tsuk-turn rabJ, DNO = 1, PNO = 3, o-turn rabcJ, 3. The combination/gram with information regarding the extraction of the individual pieces is sent to the identification unit 4. Next, the process is repeated with DNO=2 except for the first point sequence block "a", and the pattern "l)" with DNO=2 and PNO=1, DNO
= 2, the identification unit 14
1)N except for the cluster of points at the beginning [1]
The process is repeated as O-2, and A-turn rcJ with DNO=3 and PNO=1 is sent to the identification unit 14.

The character determination unit 15 selects a certain interval α based on the information regarding cutting out.
If the tsukutan within is considered to be an individual tscutan of one character, the identification result is that one! If the pattern is considered to be multiple patterns, then from among the identification results of each of the multiple combination patterns, excluding the jet, among the combination patterns, the turn width is 4 square meters. The longest one is output as the correct identification result, and subsequent identification results that include a cluster of point sequences within the section within each turn are excluded.

FIG. 6 is a flowchart showing the details of the processing in the character determining unit 15, in which information regarding the cutting L7/e turn, that is, the operation number]) is used to determine the target section for outputting as a reading result from NO and the pattern number PNO. It is determined whether or not it is a combination of patterns, the identification result is stored in a buffer, and those that can be identified based on the employee's combination pattern are output as reading results from the stored buffer.

Next, the process of character extraction and character determination will be explained using the line turn 20 of the mlr diagram as an example. For the tsukturns ``be'', ``riJ, r l-J'' in the row and o-turns 20, the number of clusters of point sequences within a certain interval α in the point sequence data 30 is one, so each - character The identification result is sent to the cutout output terminal 6. A certain section α (herein referred to as target section 6) that includes the next cut turn "ru" is cut out as an individual cut turn 21.
Since there are two clusters of dot sequences, the character cutting part 13i
dIndividual chisel turns that sequentially combine the two patterns
The identification unit 14 collects information regarding ``ノ'' and ``ru'' and their extraction.
At the same time, the position excluding the first block ``NO'' of the point sequence block U in the target section (2) is set as the reference position of the next target section (2). Here, a cluster of two point sequences is also detected in the target section (■), and information regarding the combination pattern and its extraction is sent out in the same way as above.
The same applies to the lower target sections ■ and ■. The identification unit 14 distinguishes “No” and “1” from the pattern “No” in the target section ■.
It outputs characters such as "ru" as the recognition result, and outputs the character "ru" for the pattern "ru". The character determining unit 15 outputs "ru" as the reading result in the target section ■, which has the widest pattern width and has the highest accuracy of identification result, and at the same time eliminates the identification result of the target section ■ that includes the pattern "shi". , the next character is determined based on the identification result of the target section ■. From the identification result of the target section (2), the character "C" is output as a reading result, and the next section (2) is excluded. The following pattern
of".

Regarding ``Shin'' and the like, the ``l'' is taken as a single character in the same way as above. FIG. 7 shows how the line pattern cutout, identification, and character determination processing of FIG. 4 explained above is executed, and FIG. 8 shows the flow of the processing.

In this way, according to the above embodiment, it is possible to distinguish between a single character pattern and a non-character pattern based on the number of clusters of point sequences within a certain interval σ. It is possible to reliably distinguish between the sections that should constitute the matching pattern, and if a cluster of one or more point sequences exists within a certain interval α, the position excluding the first cluster can be used to Since it is set as a reference position, all possible combination patterns can be extracted, and reading accuracy can be improved. Since the character cutting section only needs to mechanically cut out ξ turns according to the number of clusters of 9 points, there is no need to wait for the identification result of the auxiliary identification section as in the conventional case, and this entire process is configured as a pipeline. You can also speed up the processing.

As explained above, according to the present invention, in a character reading method in which character recognition is performed by cutting out each character from the chisel turn of a character line obtained by scanning and photoelectrically converting a text on a book, a predetermined Check the number of clusters of point sequences that exist within a certain interval, and if there are - pieces, consider that interval as a pattern of one character and cut it out, and if there are a number of digits, combine the clusters of the point sequence sequentially as appropriate. A cutting step in which each of the plurality of combination patterns is regarded as a pattern of one character and is cut out, and the cutout pattern and information related to the cutout are outputted; If it is regarded as a pattern, the identification result is output as is, and if it is recognized as a pattern, the set with the longest pattern width is output from among the identification results of the multiple combinations of nomiturns. In order to perform a character determination process that outputs identification results corresponding to matching patterns, it is necessary to perform complex identification and judgment on character extraction from a document that includes separated characters and cut characters and whose character pitch is not constant. It can be done unambiguously, speeding up processing, and even if some of the characters are similar to other characters, they can be read correctly.
In addition, when a plurality of clusters of point sequences exist within a certain interval, the number of consecutive clusters of dot sequences is sequentially increased by - pieces, and the combined chimiturns are regarded as one character pattern and cut out, and the plurality of clusters are By setting the position excluding the first cluster of points in the cluster as the reference position for the next certain section, all possible combination patterns can be extracted and reading accuracy can be improved. This has advantages such as being able to expand the range of objects to be read.

[Brief explanation of the drawing]

The drawings are for explaining the present invention, and FIG. 1 is a block diagram showing a conventional character reading device, FIG. 2 is a block diagram showing another conventional character reading device, and FIG. 3 is a block diagram showing a conventional character reading device. A block diagram showing an example of a character reading device according to the present invention; FIG.
5 is a flowchart of the character extraction section 13, FIG. 6 is a flowchart of the character determination section 15, and FIG. 7 is an explanatory diagram showing how the extraction, identification, and character determination processing is executed for the line pattern of FIG. 4. FIG. 8 is an explanatory diagram showing the flow of the process shown in FIG. 7. 11... Input terminal, 12... Pattern memory, 13
...Character cutting section, 14...Identification section, 15...Character determination section, 16...Output terminal Patent applicant Nippon Telegraph and Telephone Public Corporation agent Patent attorney Yoshi 1) Takashi Sei Figure 1 Figure 2 Figure 3 Figure 4 379- Figure 5 - J-R Sannosho (spontaneous) March 7, 1980 Director-General of the Patent Office Kazuo Wakasugi 1 Display of case 1989 Patent Application No. 222489 2 Invention Relationship with the name character reading method 3 amendment case Patent applicant address 1-1-6 Uchisaiwai-cho, Chiyoda-ku, Tokyo Name
(422) Nippon Telegraph and Telephone Public Corporation Representative Tsune Shindo 4 Agent 105 Telephone (03150B-9866)
Contents of 7th Amendment (1) Figure 5 of the drawings will be amended as shown in the attached sheet.

Claims (2)

[Claims] (1) In a character reading method that performs character recognition by cutting out each character from an O-turn in a character line obtained by scanning and photoelectrically converting text on a document, points that exist within a predetermined interval on the character line Check the number of clusters in the sequence, and if there are -, consider the interval as a single character pattern and cut it out, and if there are multiple clusters, create multiple combinations of nomiturns by suitably combining the clusters in the sequence. A cutting step in which the chisel turn is regarded as one character and is cut out, and the cut out chisel turn and information about the cut out are outputted; and the identification result of the cut out chisel turn and the information about the cut out, if it is considered to be one character chisel turn, the identification. A character determination step that outputs the result as it is, and if it is considered to be a plurality of patterns, outputs the identification result corresponding to the combination nomiturn with the longest pattern width from among the identification results of each of the plurality of combination patterns. A character reading method characterized by the use of . (2) In a character reading method that performs character recognition by cutting out each character from the O-turn in a character line obtained by scanning and photoelectrically converting text on a form, the characters that exist within a predetermined interval on the character line Check the number of clusters of point sequences, and if there are - pieces, consider that section as a chisel turn for one character and cut it out, and if there are multiple clusters, sequentially increase the clusters of consecutive point sequences by - pieces and combine them to create a chisel turn. Each of them is treated as a chisel turn of one character and cut out, and the position excluding the first cluster of the plurality of point sequences is used as the reference position of the next certain section,
A cutting step that outputs the cutout, o-turn, and information regarding the cutting, and the identification result of the cutout chisel turn and the information regarding the cutting, and if it is considered to be a chisel turn of one character, the identification result is output as is. However, if a plurality of chisel turns are considered, the character determination step outputs the identification result corresponding to the combination chisel turn with the longest pattern width among the identification results of each of the plurality of combination chisel turns. A character reading method characterized by 5 things.