JPS58154899A - Word recognition equipment - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION According to the present invention, the character string output from the character introduction (oci) is given in advance and can be predicted within nine words, and its range is often limited. In such a case, even if a few characters in Ka Tsumugi O cannot be read, if you use Simple-1 as a character string, the word can be determined from the entire OCR output character string, and the word can be recognized. rate can be significantly improved.

Now, in the conventional O unit knowledge, it is possible to specify one category in the OC'& output string using OCR.
If there is a character that outputs two or more possible categories in parallel, create a seven-string string for each of those power category changes, perform a simple MaN weave for each string, and then The method used was to obtain the final solution by integrating the results of *
There were 5 songs that I wanted to spend a lot of time on.

The purpose of the present invention is to solve this problem, and to make it possible to include characters in one base word that cannot be specified in one category (even if there are characters, all characters cannot be specified in one category).
The object of the present invention is to provide a device capable of hoeing at both speeds at the same speed as in the case of nine.

According to the present invention, the literary magazine wk device sequentially outputs,
lf or multiple V magazine Ik for each character that makes up ***
A plurality of character string storage devices each storing the results as a plurality of character strings, and a nine-word dictionary discipline device storing a plurality of pre-selected words.

a word storage device for temporarily storing words retrieved one by one from the word dictionary storage device; and a word storage device for temporarily storing words retrieved one by one from the word dictionary storage device; Means for determining the degree of difference between characters using nine pre-memorized characters and outputting the degree of difference in character amount for each character string, means for selecting a difference number, a single candy relation difference vertical calculation '+ stage for adding the minimum character amount difference degree for each word and calculating a word relation difference length, and each word in the single connection t11111 storage device; There is a word recognition device having means for detecting the minimum among the single candy phase j1 degrees for - and outputting the single word with the smallest word difference quality as a ml1lIa spring.

The present invention will be explained in detail below with reference to wWIi.

FIG. 1 shows data showing an embodiment of the present invention.As for the 6-pot knowledge result, 1 may be a plural number of 2 or more, but 11* is better for explanation t-@&. 2 will be explained below. In the first wA, each character constituting the word is embroidered in the embroidery device 4, and the results are outputted to the memories 5 and 7, where they are stored as the embroidery characters i. When there is only one recognition result, the m- category is stored in the memories 5 and 7, but the character-weaving device cannot specify the category v- of 5'''), and the two categories are output in parallel. In this case, they are stored separately in memories 5 and 7. In the case where two character line methods are used simultaneously, the MmM results for each method are stored separately in memories 5 and 7. In this way, the OCR recognition results for one rate 11 are stored in memories 5 and 7. JIL candy has been used in advance as d!: single-connection L memory 2, which is taken out one by one from here.
ni- ■ Memory vomits. In Lenstar 3, one character taken out sequentially from the letter 1 stored in Memory 2 is stored.The Lenstar base 8V (is 1 1k WL which is taken out into Lenster 3 and corresponds to the nine-pointed character.) Memory 5 One character from the character string stored in and 7 is retrieved and stored. 6 Memory 9 and 10 store the character amount difference in the form of a table in advance, and the character string corresponding to the two characters When a code is given as an address, it can be obtained as a sequence of letters or as an output from memory. , it is small for letters with similar shapes, and it is 0 between the same letters like E and E.
It is. The comparator 12 determines the magnitude of the inter-character phase difference output from the memory 9 and the memory 10, and the selector 11 selects the smaller character amount difference according to the output of the comparator 12 and applies it to the adder 12. Output to 13. The register 14 stores the intermediate results of the inter-word dissimilarity, and this content and the new inter-character phase! and are added in an adder 13, and its output refreshes this content. Phase 1IAII with one single fetched into memory 2! When the calculation is completed, the register 14 stores the #i pardon/end O single word dissimilarity, and this value and the previous minimum interword dissimilarity stored in the register 16 are compared to the comparator 15. and register 14
If 0 content is smaller, compare I! 15, the contents of the register 14 are transferred to the register 16, and at the same time, the semi-fine data stored in the memory 2 is transferred to the memory 17. The result is sent to output 10G.

If the contents of the register 14 are inu-mei, no transfer is performed.Then, 5K is stored in the word dictionary storage device 1, and the words 1 and 1 are stored in the memories 5 and 7. When the comparison with two character strings is completed, the memo jj17 stores the single word that obtained the minimum dissimilarity between words, that is, the recognition result,
The register 16 stores the degree of dissimilarity between the single word and the character string output from the JCB.
If there are three or more OCR #151m results, it is clear that the same result can be achieved by increasing the number of memories 7, registers 8, and memories 100 according to the number.

As explained in detail above, base #hM4 according to the present invention
No matter how many characters there are in one single category that cannot be classified into one category by OCR, they can be identified as quickly as if all the characters could be identified in one category. 4L pongee recognition can be done quickly.

We will briefly describe the case where a calculation device using the Drama and Nang method is used instead of the word difference 1 needle calculation system 18 shown by the dotted line in Figure 1. The second figure is a conceptual diagram shown to explain the method of calculating the degree of dissimilarity between words using the DP machining method.
It is assumed that BitFN:l is output. This is an example where the single #h from the dictionary is set as (BBRGiiiN). In addition,! represents characters that are unrecognizable from 0CRK. In the table, the circled numbers are the degree of similarity between characters, and the number to the left is the intermediate result of the degree of dissimilarity between words. As shown in FIG. 3, this intermediate result is as shown in FIG.
, the smallest of the upper left, top, and left intermediate results in ,
It can be found by adding the difference between characters @ in 11111VC. That is, as in kite 3-, when the upper left of the second @ is a person, the top is B, and the left tC, and the intermediate result of the desired position is X, it is determined as X = MIN (person, B, C) + @. The direction of the arrow in FIG. 2 is No. 111 for obtaining intermediate results.

FIG. 4 is a block diagram showing an embodiment of a complete unit @ difference storm calculation device using the DP Ma-Nang method. Note that in this embodiment, the number of collation machines was set to 2. In Fig. 4, when the display width n is 2, character matching requires intermediate results of two characters before and after the center, and five registers 21 to 25 are checked. Preparation (7) K is shown in the middle of the difference between words, bounded by !11F(t).Comparator 27ri compares the contents of registers 24 and 25, and selects the smaller one using selector 2ft according to the result. The - and the contents of the register 21 are compared by the comparator 31, and according to the result, the smaller one is selected by the selector 30.
, B, CK corresponds to selector 300 output 43ij
When obtaining the intermediate result at the left end of each row of 6 in FIG. 2, which is the minimum value, since there is no - in C, when the loo signal 33 is generated, that signal is supplied to the selector 30 through the 0 legut 32, The output of the selector 26 is forcibly selected. When obtaining the intermediate result at the right end of each row in Figure 2, since there is no - in B, when the right end signal 29 is generated, that signal is supplied to the selector 26 through OA'& gate 28, and is forced to register. 25 - is selected.

The input 19 of the addition@34 is the output of the selector 11 in FIG.

In this embodiment, the tlJ cabinet is set to 2, so in the selector 36, if a single connect character is lost while obtaining the intermediate result of 5 (b), as in the case of the result shown in Figure 2, the register 250 Select the output and other outputs
7 is occurring and selects the adder 340 output. The manual input in registers 21 to 25 is at the same time Ktff
be exposed. In E-11, writing 4 characters into register 3 from 2 characters is done 51 m per % when one character sound is written in register 6.8, and the input is 51 m with the arrow 2-. It is. At 9 o'clock when the calculation of the difference 1 of the semi-pongukaku is completed, the five final results are stored in the registers 21-25.

In the case of 95 cases in FIG. 2, it is indicated by the headquarters in the lower right corner and corresponds to 95 values. The selector 38 selects these registers 21-
250One out of five outputs to signal 39! , but which output to use depends on how the OCR output string is handled.
and the length M of the beginning extracted from the word dictionary, (
The value of N-M) is -2, -1, 0, +1. +2 selects the outputs of registers 21, 22, 23, 24, and 25, respectively. That is, all characters in the 9-character string and single conjunction are evaluated, f
Select the final result at position c. This value corresponds to the degree of difference between the nests tIb when the DP massing method t#+4 is used. Even if there is an increase or decrease in the number of characters composing the jIL candy, it is possible to correctly recognize the jIL candy.

[Brief explanation of the drawing]

Fig. 1 shows an embodiment of the present invention; , a conceptual diagram shown to explain the calculation method of intermediate results at each position in Fig. 2, and Fig. 4 is a conceptual diagram showing an example of a word relation dissimilarity calculation device using the DP Manning method. It is a diagram. In prison, l#i single-link dictionary placement device is character line tR
device, 2, 5.7 is memory, 3, 6.8 is register, 9
．． 10Fi character difference 11 is memorized, 12.1
5 is an IT textbook, 11 is a selector, 13 is a door calculator, 14.
16 registers, 17 memo change, 1B#i rate-at
j! ! l it calculation device, 19 and 20 are 18 human power and output, 21-25#i Renstar, 26゜ao, a6.3g is selector, 27.31Fi comparator, 34ed addition [t, 2
ji, 32dOB, y-). 21, 33, and 37 are signals, 39 is a select signal of the selector 38, and 43 is an output of the selector 3o.

Claims (1)

[Scope of Claims] 1. A plurality of character string storage devices each storing m number of recognition results for each character constituting a word, which are sequentially output from a character line processing device, as multiple aO character strings. and,
A Kudan candy dictionary storage device that stores a plurality of words that can be selected in advance, a single ME storage device that temporarily stores nine words that are taken out one by one from the Mukudan Tsumugi dictionary storage device, and a Scissor Danron discipline device. means for determining the difference between each character and a corresponding character of tK in the plurality of character string storage devices using nine characters stored in advance, and outputting a character amount difference for each character string; Means for selecting the smallest character difference number from the character difference wrinkles of each character string, and calculating the difference between words by adding the character amount difference Ill for each character string. a degree calculating device; a means for detecting the smallest one among the degree of dissimilarity between words for the six units in the single dictionary storage device and outputting the word with the minimum degree of dissimilarity between words as a recognition result; A single Mal weaving device characterized by having the following. 2. Dr. Ma as a calculation specialist for the degree of difference between the above-mentioned single pongmugi. In the process of storing intermediate results of the number of word dissimilarities that are equal to the value obtained by adding 1 to twice the matching width in the matching method, the #next memorization star, one character in the character string storage device, and this Calculate the degree of difference in character amount between the character #&, which is equal to the sum of twice Kl of the collation -, and centering on the character in the stroke Ink single lume recording device located at the position corresponding to the character. , means for calculating an intermediate result of a new inter-word dissimilarity degree according to the kDP machining method using these values and the value stored in the intermediate result storage register, and storing it in the intermediate result storage register; Phase j with two words! When the degree calculation is completed, one of the outputs of the intermediate result storage regime is determined by the difference between the number of characters in the character string and the number of characters in the word from the vertical writing.
2. A weaving apparatus according to claim 1, further comprising means for selecting one word and outputting it as a degree of difference between words.