JPS63268083A - Word recognizing device - Google Patents

Abstract

PURPOSE:To easily detect a relevant word despite the misrecognition of a candidate character string by retrieving an area and its periphery in a word dictionary in response to the number of characters and the degree of complicatedness of the candidate character string. CONSTITUTION:A degree of complexity table 7 defines the degree of complexity of each character. A word dictionary 9 stores words in the order of the number of characters and the degrees of complexity of words. An area indication table 8 stores area information on the area in the dictionary 9 where the words are stored in response to the number of characters and the degrees of complexity. A recognizing part 4 refers to a recognition dictionary 5 for the characters extracted by a character extracting part 3 and selects the corresponding candidate character string for each character. A word collating part 6 refers to the table 7 for calculation of the degree of complexity of the candidate character string and reads the area information out of the table 8. Then the part 6 collates the words read out of the corresponding area and its periphery in the dictionary 9 based on the area information with the relevant candidate character string to detect the relevant word.

Description

[Detailed Description of the Invention] [Summary] This invention provides a word recognition device that recognizes and outputs a word corresponding to a candidate character string. In order to solve the problem of not being able to find the correct word by sequentially searching the area starting from the first character of this misrecognized word in the word dictionary, we have created a word dictionary that stores words in order of number of characters and complexity. By searching the area in the word dictionary and its vicinity corresponding to the number of characters and complexity of the candidate string to find the correct word, even if there is a minor misrecognition in the candidate string, it can be easily detected. I'm trying to find the correct word.

[Industrial application field]

The present invention is configured to provide a word dictionary in which words are stored in order of number of characters and degree of complexity, and to search an area in the word dictionary corresponding to the number of characters and degree of complexity of a candidate character string and its vicinity to find the corresponding correct word. This invention relates to a word recognition device.

[Prior art and problems to be solved by the invention] Some character recognition devices have a word dictionary and use this to improve the word recognition rate. A brief explanation will be given below with reference to FIGS. 7 and 8.

The characters written on the paper, such as "automobile", are read using a scanner or the like, and each character is recognized to generate candidate character strings as shown in FIG.

This generated candidate character string is compared with a word dictionary as shown in FIG. 8, and a matching string or, if there is no matching string, a string with a high degree of similarity is output as a recognition result. At this time, the conventional word dictionary shown in Figure 8 is
Words were arranged in order of "number of characters" and "character code". Therefore, the candidate character string at the beginning of the handwritten character is recognized as "self", and in the case of a three-letter word, the corresponding correct word "" is easily recognized as shown in Figure 7.
car” can be found.

However, as shown in the first to third candidates in Figure 7, for some reason the first character "self" in the candidate character string is mistakenly changed to "eye", "white", or "day". When the words are recognized, there is a problem in that a word consisting of three letters including these letters at the beginning cannot be found even if the word is searched from the 8th round word dictionary.

[Means for solving problems]

In order to solve the above problems, the present invention includes a character recognition unit 4 that recognizes each input character and generates a candidate character string, a complexity table 7 that defines the complexity of characters, and a character recognition unit 4 that recognizes each input character and generates a candidate character string. A word dictionary 9 storing words in order of word complexity, and an area instruction table 8 storing area information indicating in which area in the word dictionary 9 words corresponding to the number of characters and complexity of a candidate character string are stored. Based on the number of characters in the candidate character string and the complexity of the candidate character string calculated with reference to the complexity table (7), the corresponding area in the dictionary 9 is determined based on the area information read from the area instruction table 8. A word matching section 6 is provided to find the corresponding word by matching the word read from the area and its vicinity with the candidate character string, and the word found by the word matching section 6 is output.

FIG. 1 shows a basic configuration diagram of the present invention. Observation section 1 in the figure is
A scanner or other device is used to read forms containing handwritten characters.

The control unit 2 performs various controls.

The character feature extraction section 3 extracts character features from the binary image signal read by the observation section 1.

The character recognition unit 4 generates candidate characters for each character by referring to the recognition dictionary 5 based on the character features extracted by the character feature extraction unit 3.

The word matching unit 6 searches the word dictionary 9 for words that correspond to the candidate character string recognized by the character recognition unit 4.

The complexity table 7 stores complexity levels for characters.

The area designation table 8 stores area information in which words corresponding to the number of characters in a candidate character string and the degree of complexity calculated using the complexity table 7 are stored in the word dictionary 9.

The word dictionary 9 stores words in order of number of characters and degree of complexity.

The output unit 10 outputs words found in the word dictionary 9 by the word recognition unit 6 as character recognition results.

[Effect]

Next, the operation will be explained.

In FIG. 1, when a form with handwritten characters, etc. written thereon is input into the observation unit 1, the handwritten characters, etc. are read and notified to the character feature extraction unit via the control unit 2. The character feature extraction unit 3 extracts character features. Character recognition section 4
refers to the recognition dictionary 5 and selects corresponding candidate characters for each character based on the characteristics of the extracted characters. The word matching unit 6, which has been notified of these IF-translated candidate character strings, refers to the complexity table 7 to calculate the complexity of the candidate character string, such as the total number of strokes, and calculates the number of characters in the candidate character string and the calculated number of characters. Area information in which words corresponding to the degree of complexity are stored in the word dictionary 9 is read out from the area instruction table 8. This read area information f [i4 corresponding word dictionary 9
Words are sequentially read from the inner region and its vicinity and compared with candidate character strings.

As a result of the matching, a matching word, or if there is no matching word, a word with a high degree of similarity is outputted via the output unit 10 as a recognition result.

As described above, a word dictionary 9 in which words are arranged in order of number of characters and complexity is provided, and an area and its vicinity in the word dictionary 9 in which words corresponding to the number of characters and complexity of a candidate character string are stored are searched. By comparing the handwritten characters and finding the correct word, even if there is a small misrecognition in the candidate string of handwritten characters, such as a character with a slightly different number of strokes, it is possible to find the correct word, improving the recognition rate. be able to.

〔Example〕

Next, the configuration and operation of one embodiment of the present invention will be explained in detail using FIGS. 2 to 6.

FIG. 2(a) shows an example of the word "automobile" written in the form. This is a form with handwritten characters written on it.

FIG. 2(b) shows candidate character strings. This was generated as follows. The handwritten character "car" shown in FIG. 2(a) is read by the observation unit 1 in FIG. A character feature extraction unit 3 extracts character features from this read binary image. Based on the extracted character features, the character recognition unit 4 refers to the recognition dictionary 5 and generates candidate character strings as first to third candidates in descending order of similarity.

FIG. 3 shows an example complexity table. This represents an example of the contents stored in the complexity table 7 in FIG. 1, and represents the complexity of a character by, for example, the number of strokes. When searching for the candidate character string, for example, the first candidate "car" shown in FIG. 2, with reference to this complexity table 7, the character "eye" has a complexity of "5", and the character "movement" has a complexity of "11". Since the character "car" has a complexity level of "7", the complexity level for the candidate character string of the first candidate is calculated as "23". Note that the word "automobile" is calculated to have the number of characters "3" and the complexity level "24".

FIG. 4 shows an example of a word dictionary in which words are stored in order of number of characters and complexity. This shows an example of contents stored in the first colloquial language dictionary 9. In this word dictionary 9, the words are arranged in order of the number of characters in the words and the degree of complexity of the words calculated using FIG. Specifically, first, the words are arranged in order from the least complex one character to the highest complexity. Second
The two-letter words are arranged in order from the least complex to the most complex. Arrange in the same manner below. This means that for the number of characters handwritten on a form, it is almost certain that i = R.
Therefore, this number of characters is stored with first priority. As for the complexity of each character, there is a high possibility that it will be misrecognized, so if this is recognized with a slightly different value and a candidate character string is generated, °ζ will also make it easier to search for the correct word. They are arranged and stored in order of complexity (for example, number of strokes). In this way, the number of characters and ? ! [By providing a word dictionary 9 in which words are arranged in order of complexity, even if a candidate character string with a slightly incorrect number of strokes is generated as shown in Figure 2 (b), the correct word can be easily collated. It becomes possible to find it.

FIG. 5 shows an example of the area designation table. This represents an example of the contents stored in the area instruction table 8 in FIG. In the figure, the first column from the left, "Number of characters," represents the number of characters in the candidate character string, the second column, "Complexity," represents the complexity of the candidate character string, and the third column, "First character number." represents the number at the beginning of the word stored in the word dictionary 9 that corresponds to the number of characters in the first column and the complexity level in the second column, and the "end character number" in the fourth column represents the number of words corresponding to the number of characters in the first column and the complexity level in the second column. represents the end (end) number stored in the word dictionary 9. This area designation table 9 is provided, and the number of characters in the candidate character string and the number of characters in the candidate character string are shown in FIG.
! The starting number and ending number in the word dictionary 9 that stores words corresponding to the complexity of the candidate character string obtained using the complexity table 7 are obtained, and matching is performed only with words having the corresponding number of characters and complexity. It becomes possible to do so. In addition, if the correct word is not found within this complexity level, read out the start number and end number of the area in the word dictionary 9 whose complexity level is one more or one less, and perform further comparison.
It becomes possible to find the correct word corresponding to a candidate character string recognized with a slightly different number of strokes.

Next, the operation of the configuration shown in FIG. 1 will be explained in detail using FIG. 6.

In FIG. 6, ■ indicates a character recognition state.

The character recognition unit 4 in FIG. 1 refers to the recognition dictionary 5 to generate a candidate character string, for example, the first candidate or the first candidate shown in FIG. This means generating a third candidate character string.

In the figure, ■ indicates a state in which the complexity level C is calculated with reference to the complexity level table 7. This means that, as explained using Figure 3, the complexity is calculated for each character in the candidate character string, and then the total complexity is calculated, for example, the complexity of "rotating wheel" is set as "23". are doing.

In the figure, ■ indicates a state in which the corresponding area in the word dictionary 9 is determined by referring to the area designation table 8. This is because, as explained using FIG. 5, there are words in the word dictionary 9 that correspond to the number of characters and complexity of the candidate character string, for example, the number of characters in the candidate character "turning wheel" is "3" and the complexity is "23". This means searching and finding the stored starting and ending numbers.

■ in the figure indicates a state in which the word dictionary 9 is referred to.

This is a word corresponding to the number of characters and complexity of the candidate character string in the word dictionary 9 shown in FIG. This means that the words in the area where the vulgar words corresponding to `` are stored are sequentially read out and compared with candidate character strings.

In the figure, ■ indicates a state in which it is determined whether a word has been spelled or not.

This means that the words sequentially read out from the corresponding area in the figure are compared with candidate character strings to determine whether there is a match or a highly similar word. . In the case of YES, the result is outputted as a recognition result, for example, printed out, as indicated by ■ in the figure. In the case of NO, set the value of complexity C to “+1” and “-1” at ■ in the figure.
Then perform the following steps (■ in the figure) to match words that correspond to the complexity level in the vicinity. As a result, even if a handwritten character is read (recognized) with a slightly incorrect number of strokes, the correct word can be found and output, and the word recognition rate can be improved.

〔Effect of the invention〕

As explained above, according to the present invention, a word dictionary is provided in which words are stored in order of the number of characters and complexity of the word, and an area in the word dictionary corresponding to the number of characters and complexity of a candidate character string and its vicinity are searched. Since the system uses a configuration to find the correct word using a combination of words, even if there is a small misrecognition such as the number of strokes in the candidate character string, it is possible to easily find the correct vulgar word, which improves the word recognition rate. be able to.

[Brief explanation of drawings]

Fig. 1 is a diagram showing the basic structure of the present invention, Fig. 2 is an explanatory diagram of candidate character strings, Fig. 3 is an example of a complexity table, Fig. 4 is an example of the contents of a word dictionary, Fig. 5 is an example of an area specification table, FIG. 6 is an explanatory diagram of the operation of the present invention, FIG. 7 is an example of candidate characters, and FIG. 8 is an example of the contents of a conventional word dictionary. In the figure, 4 represents a character recognition unit, 6 represents a word matching unit, 7 represents a complexity table, 8 represents an area instruction table, and 9 represents a word dictionary.

Claims (1)

[Claims] A word recognition device that recognizes and outputs a word corresponding to a candidate character string, comprising: a character recognition unit (4) that recognizes each input character and generates a candidate character string; A complexity table (7) that defines the degree of complexity, a word dictionary (9) that stores words in order of the number of characters in the word and the complexity of the word, and a word dictionary (9) that stores words that correspond to the number of characters and complexity of the candidate string. The complexity of the candidate character string calculated by referring to the area instruction table (8) that stores the area information of which area it is stored in, the number of characters in the candidate character string, and the complexity table (7). The dictionary (
9) a word matching unit (6) that matches the candidate character string with the word read from the corresponding area and its vicinity in the area to find the corresponding word; A word recognition device characterized in that it is configured to output words.