JPS58132871A - Optical character reading device - Google Patents

Abstract

PURPOSE:To improve the reading accuracy for an input character pattern, by retrieving and recognizing the input character pattern which was not recognized at the 1st discriminating part from the contents of a modified character dictionary part which is closest to a standard character pattern through the 2nd discriminating part. CONSTITUTION:The 1st discriminating part 5 supplies an answer code corresponding to the standard character pattern of the highest resemblance to an answer editing part 8 by giving consideration to the difference between the highest and the next highest resemblances among standard character patterns to the input character pattern which is obtained from an operation of resemblance. At the same time, the part 5 supplies the answer code corresponding to the 1st candidate character pattern to the 2nd discriminating part 7 in addition to the input character pattern. The part 7 performs an operation of resemblance between the input character pattern and all read-out modified character patterns to decide that the modified character patterns which have resemblances higher than a certain standard to the input character pattern are recognized.

Description

[Detailed description of the invention] (Technical field of the invention) This J! Ming is an optical character m loading device (hereinafter referred to as OC 凰) K
This is related to p1 and ocaK, which can also recognize modified characters.

[Invention O technical background and its problems]

Currently, the character R in 000H has a 1lillt phlegm-like character type that gradually changes from mechanical characters to I! K is good so that you can take it.

By the way, in OCRog knowledge 1, a standard character pattern is prepared in advance, and it is input and compared with the round character pattern 24m1 quasi-character pattern, and it is determined whether or not a similar fK of a certain value or more is achieved. A common method is to determine the character based on the character. In other words, even if it is a regular handwritten character, the regular handwritten character that can be read by OC 几 is
Subject to Kanapu restrictions.

As a result, so-called modified characters that deviate slightly from standard character patterns but are easily readable by humans are often rejected or misread.

On the other hand, in order to improve the accuracy of the %vt removal rate, there is a method in which the character pattern is once input to the recognition section, the rejected character pattern is logically changed to the grayscale slice level t-, and then the rejected character pattern is input to the recognition section for recognition. However, with such a method, it would take an extra amount of time to recognize the h degrees. Also, for modified characters, even if you change the slice level, wtMi
There are many cases where I can't do anything.

[Purpose of the invention]

The present invention aims to improve reading accuracy regarding modified characters, and
m! The purpose of the present invention is to provide an OCR that performs processing m in a short time.

[Outline of invention cost]

In OCR, the present invention includes a standard character dictionary section, a first identification section that identifies an input character pattern based on the contents of the standard character dictionary section, a modified character dictionary section, and a rattan 2 identification section. The identification unit of the wi2 searches the contents of the modified character dictionary unit for the input character pattern not recognized by the first identification unit based on the code of the standard character pattern closest to the input character pattern, Let your friends recognize you in a short time.

[Practice of the invention]

The following is a review of one*m example of this invention with reference to the drawings.
14 Suishi, 111 Gang ((Ten) Hiroshihoji (explanatory diagram showing an example of the character), 1111 (b! is! An explanatory diagram showing an example of the R-shaped character. In the irises 1 diagram (b), the dotted circle 8 in the bar, 9 parts, off the point of the Fi standard character, 9 parts.Second
＠ is a block diagram showing one embodiment of the present invention. In the figure, 1 is a photoelectric transformer l1lIIWA. The photoelectric conversion unit 1 has a function of converting circle characters written on a form into electrical character pattern signals. 2 is pretreatment part m, pretreatment part 2
The character pattern is shaped like Il and the noise is removed.
is, Toi, and 3 are twin buffer sections. Twin buffer vs. 3-line, which stores rounded character patterns on a line-by-line basis.

4 is a standard character dictionary section, and the standard character dictionary s4 stores only standard character patterns. 5 is a first identification part. The first job classification s5 inputs a character pattern for one character detected from the line buffer 3 and performs similarity calculation with all standard character patterns stored in the standard character dictionary s4. As a result of the calculation, if there is a standard character pattern that satisfies the degree of similarity above a certain standard, an answer code corresponding to that standard character pattern is output; if there is no standard character pattern that satisfies the degree of similarity above a certain standard, then
Together with the input character pattern, an answer code corresponding to the standard character pattern (hereinafter referred to as Ill candidate character pattern) having the highest degree of similarity to the input character pattern is output. 6 is a modified character dictionary section. In the modified character dictionary s6, one or more modified character patterns corresponding to each standard character pattern stored in the standard character dictionary s4 are classified and stored. FIG. 3 shows the contents of the modified character dictionary section 6, and shows the specifics of the number [2]K in %? 11 is shown. 7 is an identification part of jI2. The second identification unit 7 is based on the irises ICI identification 1i15,
When an answer code corresponding to the first candidate character pattern is supplied together with the input character pattern, the answer code corresponding to this standard character pattern is set as an address in the modified character dictionary 6, and the modified character stored at this address is Only the pattern is generated, and the similarity calculation between this modified character pattern and the input character pattern is performed. As a result, if there is a modified character pattern that satisfies the degree of similarity equal to or higher than a certain standard, it corresponds to the standard character pattern. If there is no modified character pattern that satisfies the degree of similarity equal to or higher than the predetermined standard, the answer is output. 8 is the answer editing department, and answer editing s8 runs all the irons 115 and 111.
It is for editing the answer supplied by the identifying section 7 of 12.

Next, the operation of the device developed by the present invention will be explained in detail.

The rounded characters entered in the notebook 11 are converted into electrical character patterns in the photoelectric conversion section I. This character pattern is formatted in the pre-processing area 112, paper noise etc. are removed, and after rounding, it is input to the twin buffer unit 3, and the 2472177 unit 3 holds one line of character pars tli. 9 Impaqua Department 3 owns 9 ko (the character pattern for 11 lines is.

Each character is @tayamaed and supplied to 1a1's Ori Tauben 5.

The first identification s5 is a similarity calculation t- between this input character pattern and all standard character patterns stored in the standard character dictionary section 4, when a character pattern for - characters is input.
Let's hit. The identification unit 5 of 111 considers the highest degree of similarity and the second highest degree of similarity among the degrees of similarity of each standard character pattern to the input character pattern obtained as a result of the similarity degree calculation, and selects the highest degree of similarity. If the degree of similarity is higher than the standard, this input character pattern is considered to have been recognized, and an answer code corresponding to the standard character pattern having a high degree of 1Ik4 similarity is supplied to the answer editing section 8.

When there is no standard character pattern whose degree of similarity reaches a certain standard, the identification unit 5 of %III selects j1 along with the input character pattern.
The answer code corresponding to one candidate character pattern is supplied to the identification s7 of JII2.

The second identification s7 first determines that the input character pattern that was not recognized by the identification unit 5 of the IIEI and the answer code corresponding to the first candidate character pattern for this input character pattern are supplied! Using the answer code corresponding to the 11 candidate character pattern as an address, a modified character pattern for the 1 candidate character pattern in the modified character dictionary section ε is found.

Next, the 20th identification unit 7 calculates the degree of similarity between the input character pattern and all the modified character patterns that can be read out, and determines whether there is a suitable modified character pattern whose similarity to the input character pattern is equal to or higher than a certain standard. If so, this input character pattern is recognized and is set as 9.

The identification unit 7 of the irises 2 supplies the answer code corresponding to the candidate character pattern to the answer editor Ii8.

If there is no modified character pattern whose degree of similarity reaches a certain standard, this input character pattern will not be illm'd and will be marked as 9, and the identification unit 7 of the ms will notify the answer editing unit $ of rejection.

The answer editing iIs edits the answer supplied from the irises 1 identification 115 or the tail 2 identification section 7.

The above is the identification 115 of #11 and the second identification 1i17.
This is the theory of meshing that performs ll1w & local for each input character pattern.

Next, a case in which a character pattern for one line including %-transformed characters is recognized character by character jIi 11 times consecutively will be described in detail with reference to FIG.

FIG. 4 is a timing chart showing the flow of recognition processing in the first identification section 5 and the tal identification section 7. In the example shown in this figure, the input character turn is the number roJ.
, fxJ , rzzj , faJ , r4J,
This is a case where these are sequentially supplied to the identification unit 5 of III, among which the numbers rOJ and faJ.

This is the case where "4" is a standard character and rlJ and rzz are modified characters.

First, the character pattern of the standard character "0" is the first identification fi
5, the first identification section 5 performs a recognition process on the character pattern of the standard character "0", and sends the answer, that is, the answer code corresponding to the character pattern of the standard character "0" in this case, to the answer editing section. Supply to 8.

Next, the character pattern of the first modified character “l” is detected by the first identification unit 5.
supplied to The identification unit 5 of tal recognizes the modified character ri''o character pattern after performing the recognition process mt-.9 Afterwards, the identification unit 7 of #I2 contains the character pattern of the modified character ``1'' and the iris candidate character, here the standard character `` The answer code corresponding to the character pattern of "l" is supplied. and the 20th identification s7
starts recognition processing for the modified character "l".

Next, when the character pattern of the modified character "2" is supplied to the standard character recognition section 5, the recognition section 5 of jll starts recognition processing for the character pattern of the and vfR type character "2".

Here, JP-A T1, which favors the recognition process for the character pattern of the modified character "2" in the identification ii5 of Iris, is better than the recognition process KII for the character pattern of the modified character "1" in the identification i17 of 112. The time T is also long because the first identification unit 5 uses the standard character dictionary 11
4 has to perform similarity calculations for all standard character patterns stored in @! OIll
jt41m17 Hiro 1 stored in the f-type character dictionary section 6
This is because it is sufficient to calculate the similarity of the modified character pattern 0+ for the two standard characters.

First of all, the identification part 7 of 1/s2 is the modified character "1".
'' is terminated, and the answer, that is, the answer code corresponding to the character pattern of the standard character "1" in this case, is supplied to the answer editing s8. After that, the identification s5 of the stripe 1 completes the recognition process for the character pattern of the deformed character "2", and the deformed character "2" is displayed in the identification section 7 of #I2.
and the answer code corresponding to the character pattern of the Jll candidate character, that is, the standard character "2" in this case, are supplied to the second identification unit 7. Then, the second identification unit 7
Recognition escape for the character pattern of the modified character "2" is started.

Next K, the character pattern of the standard character "3" is the identification fi of Ml
i5, and the identification s5 of this ill is the standard character “
The recognition layer is started for the character pattern "3".

Here, too, the time Tl that contributes to the recognition unit layer for the character pattern of the standard character "3" in the first recognition unit 5 is longer than the time Tl required for the recognition processing of the modified character "2" in the recognition unit 7 of $2. It's also long.

Therefore, first, the identification of 312 s7 grunt plexiform character "2"
The recognition process 11tM for the character pattern is completed, and the answer, that is, the answer code corresponding to the standard character "2" in this case, is supplied to the answer editing s8. After that, the identification s5 of the irises 1 is the character pattern of the standard character "3".
The answer, that is, the answer code corresponding to the character pattern of the standard character "3" is supplied to the answer section $IIg.

Next, the character pattern of the standard character “4” is the identification part 5 of gt.
When supplied to , Ill's identification s5 performs recognition processing on the character pattern of the standard character "4" and sends the answer code to the answer editing department 8. supply

The answer editor 11g edits the answer code supplied from the identification section 7 of the irises 10 iron section 5 or the irises 2 in the manner described above.

In this way, the character pattern for one line stored in the line buffer section 3, that is, in the above remainder, the numbers rOJ, r
Even if the lJ, rgJ, r3J, and rgo character patterns include modified characters, they can be accurately recognized in the same time as it takes for the first identification unit 5 to recognize them.

As described above, when the input character pattern is not recognized by the identification unit 5 of Ml, the identification unit 5 of IEI recognizes the standard character dictionary 1.
The first character pattern from among the standard character patterns stored in 1i4
This is an explanation of the case where a candidate character pattern is selected and the identification unit 7 of the WJ 2 selects an input character pattern tm based on the candidate character pattern. On the other hand, the first identification sb selects the second and third candidate character patterns as well as the III candidate character pattern, and the second identification unit 7 selects these first, second and third candidate character patterns. If the input character pattern is recognized based on the candidate character pattern, the accuracy of finding employment will be further improved.

〔Effect of the invention〕

As described above, by using the apparatus according to the present invention, the following effects can be obtained.

■ Reader 9 for deformed characters that are clearly readable by humans
Accuracy is greatly improved.

■ The time it takes for the recognition unit to recognize one line of character patterns containing deformed characters is greatly reduced.

Figure 1 (Jl) is an explanatory diagram showing the standard characters written on the book lK, and Figure L 1111 (The shrine is also an explanatory diagram showing the modified characters that can be written on the form. Figure 2 is a block diagram showing an embodiment of the device according to the present invention, and Figure 3 is a conceptual diagram showing the answer of the fL engraving dictionary section.

j141i is a tie chart showing the flow of recognition processing in the first identification unit 5 and the identification unit 7 of jI2.

l...Photoelectric conversion part 2...Pretreatment ring part 3...
・Line buffer s 4...Standard character dictionary section 5...
・Ill1 identification II 6... Modified character dictionary Iku 7... Second identification part 8... Answer editorial department Daizuito Patent attorney Nori Chika Kensuke (and 1 other person)

Claims (1)

[Scope of Claims] A standard character dictionary section that stores standard patterns, and identifies input unknown character patterns based on nine standard character patterns protruding from the standard character dictionary section and generates candidate characters when they cannot be identified. a first identification section that outputs a sound; a modified character dictionary section that stores modified character patterns with respect to the standard pattern; A second method for inputting a candidate character, reading out a modified character pattern corresponding to the candidate character from the modified character dictionary section, and identifying an unknown character pattern that has been input based on the modified character pattern read out intermittently. 1. An optical character Il mounting device comprising a character identification section.