JPH04372087A - English character recognition device - Google Patents

Abstract

PURPOSE:To make it possible to accurately perform the segmentation of contact characters. CONSTITUTION:After segmenting an area by a work unit from an inputted picture, an area by a character unit in a character area segment part is segmented and a recognition processing by a character unit is performed in a character recognition part. At this point, as for the character area of which recognition accuracy is low, a resegmentation is performed. In resegmentation processing, the collation of a character string of which recognition accuracy is high in the character recognition processing with a recognition dictionary with the spelling of a word stored is performed, a character string existing in an area of which recognition accuracy is low is estimated, a segmentation position is determined based on a character width dictionary storing the character width of each character and a resegmentation is performed. The character segmentation can be performed at a more correct position and the recognition accuracy can be improved by adding the information on a word dictionary and the character width dictionary at the time of resegmenting the character area.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an English character recognition device for recognizing English characters.

0002

[Background Art] In recent years, there has been an increasing demand for character recognition devices to be used as input devices for computers, etc., and character input devices that can efficiently obtain stable recognition results can improve the performance of systems such as computers. has become essential.

Conventional recognition devices perform recognition by comparing an image of a character area with a recognition dictionary. In this recognition, if the recognition accuracy is low, it is assumed that characters are in contact within the extracted character area, the area is divided equally by the estimated number of touching characters, and vertical black The recognition result was obtained by obtaining a histogram of pixels, using the position with the lowest frequency as the separation position for each character, cutting out the character area again, and performing recognition.

0004

[Problem to be Solved by the Invention] However, since the character spacing in English text is not constant, it is difficult to calculate the number of touching characters, and the width of English characters varies depending on the type of character.
Even if the area is simply divided into equal parts based on the estimated number of characters, the area will not necessarily be divided at or near character breaks. As a result, a character area may be cut out at the wrong position, resulting in a decrease in recognition accuracy. An object of the present invention is to provide an English character recognition device that solves these problems.

[0005]

[Means for Solving the Problems] In order to solve this problem, the present invention extracts character strings with high recognition accuracy as a result of recognition in a character recognition unit, and creates a character string composed of those characters in a character area extraction unit. Words are extracted from a word dictionary, and character string candidates that exist in areas with low recognition accuracy are selected. and,
The recognition accuracy is improved by estimating the character-by-character break positions of the selected character string candidates based on the character width dictionary, cutting out the character area again, and performing recognition.

[0006]

[Operation] In the present invention, when cutting out a character region with low recognition accuracy again, character string candidates are selected based on the word dictionary, and character string candidates selected based on the character width dictionary are By estimating the break position of each character, character areas can be cut out at more accurate positions to improve recognition accuracy.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described with reference to the accompanying drawings.

In FIG. 1, 1 is an image input unit that inputs a document to be recognized as a document image; 2 is a document area cutting unit that finds a collection of character strings from the input document image and outputs a document area; 3 is a document A word area cutting section 4 finds a word-by-word break from the area and outputs a range of one word as a word area; 4 is a word dictionary 7 that stores the spellings of English words; and a word dictionary 7 that stores the character width of each English character. Find character-by-character breaks from the word area based on the character width dictionary 8.
A character extraction unit 5 outputs a range of two characters as a character area, and a character extraction unit 5 compares a recognition dictionary 9 prepared in advance based on the graphical features of all characters, calculates the degree of recognition between them, and performs character recognition. A character recognition unit 6 controls the character area extraction unit 4 and the character recognition unit 5, determines a recognized character code,
A control unit that controls the output unit 10, 7 a word dictionary that stores the spellings of English words, 8 a character width dictionary that stores the character width of each English letter, and 9 a character width dictionary that stores the graphical characteristics of all characters. A recognition dictionary is prepared in advance, 10 is an output unit that outputs the determined recognition character code, 11 is an image input unit 1, a text area extraction unit 2, a word area extraction unit 3, a character area extraction unit 4, and character recognition. Internal bus connecting each part of section 5, 12,1
3 are between the character area cutting section 4 and the control section 6, respectively.
An internal bus 14 that connects the character recognition unit 5 and the control unit 6
15 and 16 are internal buses that connect the character area extraction unit 4, word dictionary 7, and character width dictionary 8, and internal buses that connect the control unit 6 and the output unit 10 and the character recognition unit 5 and the recognition dictionary 9, respectively. It is.

Regarding the English character recognition device configured as described above, FIG. 2 shows an overall flowchart, and FIG. 3 shows a character area extraction flowchart, and the operation thereof will be explained below.

First, a document to be recognized is input as a document image to the image input section. (Step s1 is simply referred to as s1. Hereinafter, the letter ``step'' will be omitted as well). The input document image is sent to the document area cutting section 2, which obtains a histogram of black pixels in the vertical and horizontal directions of the document image, and extracts the document area based on this histogram. Then, the position information of the document area is stored as internal data (s2).

[0011] Position information of the document area is sent to the word area extraction unit 3, and word area extraction processing is performed within the document area. The word area cutting unit 3 focuses on the fact that the spaces before and after a word are larger than the spaces between characters, and cuts out a character string sandwiched between spaces of a certain width or more as a word area. The position information of all the word areas in the document area found by the document area extraction unit 2 is obtained and stored as internal data (
s3).

[0012] Based on the positional information of the document area/word area obtained in steps s1 to s3 above, character area extraction and character recognition processing are performed. Cutting out the character area and recognition processing are performed for each word, and one word is recognized from s4 to s8.

Two types of flags are used in the process and have the following meanings.

[0014] The re-extracting flag indicates whether the character area is being extracted for the first time from the word area to be processed or whether the character area is being re-extracted.The value "0" indicates the state where the character area is being extracted for the first time from the word area. , the value "1" represents a state in which re-cutting is performed from the word area.

[0015] The candidate character string flag indicates whether a candidate character string exists or does not exist when re-extracting, and a value of "0" indicates a state where a candidate character string does not exist.
A value of "1" represents a state in which no candidate character string exists.

First, the re-cutting flag and candidate character string flag are set to "0". (s4).

The character area extraction unit 4 focuses on the fact that English letters are not separated to the left and right, treats the area between the front and back spaces as characters, and separates the series of characters in the word area into character units. Extract as a character area. The position information of all the character areas in the word area extracted by the word area extraction unit 3 is obtained and stored as internal data (s5 in Fig. 2, Fig. 3
s24).

The position information of the character area is sent to the character recognition unit 5, and the characters within the character area are recognized. In the character recognition section 5,
The similarity in shape between the image of the character region of interest and the character to be recognized is determined as the degree of similarity, and the recognized character code of the character to be recognized with the highest degree of similarity and the similarity value are stored as internal data. This process is performed for all character areas within the word area (s6).

The recognized character code and the degree of similarity are sent to the control unit 6, and it is determined whether character segmentation needs to be performed again. The control unit determines whether the similarity value of each character is worthy of the recognition result (s7), and if the similarity value of all the characters is worthy of the recognition result, ends the character area cutting and character recognition processing, Sends the recognized character code to the output section. If there is a character whose similarity value does not match the recognition result, it is determined whether the character area can be cut out again (s9).

In this determination, if the re-cutting flag is "0" or the candidate character string flag is "1", the position information of the character worthy of the recognition result and the recognized character code are sent to the character area cutting unit 4, and the character area is cut out again. Cut out. If the re-cutting flag is "1" and the candidate character string flag is "0", the character area cutting and character recognition processing are finished,
The recognition result is determined as a reject, and a reject code is sent to the output unit (s10). The output unit outputs the received recognized character code or reject code (s8).

[0021] The process of cutting out a character area when there are touching characters in a word area is called ``southwest''.
The case where "th" is touching is shown as an example.

[0022] In the character area cutting process, "
The characters are cut out into eight character areas: s, o, u, th, w, e, s, and t.
It is determined that "s", "o", "u", "w", "e", "s", and "t" are worthy of recognition results, but "th" is not worthy of recognition results.

The control section sends "s" and "o" to the character cutting section.
"u""*""w""e""s""t".

Each recognized character code (* indicates reject) and position information are sent to the character area cutting section 4.

The character re-cutting process will be described below with reference to the detailed flowchart of the character area cutting process shown in FIG.

It is determined whether the re-cutting flag is ``0'' or ``1'', and it is determined whether it is the first character area cropping process or the re-cutting process (s13). Furthermore, in the case of re-extracting processing, it is determined whether this is the first re-extracting process, and in the case of the first re-extracting process, the sent recognition character code and position information are stored in internal data to obtain a candidate character string. Candidate character strings are determined by comparing the sent character strings with high recognition accuracy against a word dictionary 7 that stores word spellings in advance.
A word with matching character strings is set as a candidate character string, stored in internal data, and a candidate character string flag is set to "1". In this example,
The first 3 letters are "sou" and the last 4 letters are "wes"
By comparing the word "t" with the word dictionary 7, the candidate character string "southwest" is obtained (s13, s14, s
15, s16, s17, s18).

If there is no matching word, the candidate character string flag is set to "0" and the character area extraction process ends (s23).

Next, select one of the candidate character strings, assume the number of characters and characters in the character area with low recognition accuracy, and calculate character by character based on the character width dictionary 8 that stores character widths in advance. Estimate the break position. In this example, the candidate string “sou
Assuming that two characters "t" and "h" are in contact with each other in the character area where the recognition accuracy was low, "t
” and “h” are obtained from the character width dictionary 8. (
Let these values be wt and wh. )

[0029] The break position of each character is estimated from the following formula: left end of area where recognition accuracy was low + width of area where recognition accuracy was low x wt/(wt+wh), and stored in internal data. If three or more characters are in contact, the break position of each character is similarly estimated (s19).

Obtain a histogram of black pixels in the vertical direction before and after the estimated break position, set the position with the lowest frequency as the break position for each character, cut it out as a character area again, and then Store location information as internal data. (s21).

It is determined whether a candidate character string exists, and if there are no more candidate character strings, the candidate character string flag is set to "0".
” (s22, s23).

The positional information of the newly cut out character area is sent to the character recognition unit 5, and recognition within the character area is performed.

As described above, character recognition processing for a given document image is performed by repeating steps s4 to s8 until all words in the document are recognized (s12).

[0034]

As described above, according to the present invention, it is possible to accurately determine the position where characters are in contact and to cut out a character area, thereby improving recognition accuracy.

[Brief explanation of drawings]

FIG. 1 is an overall configuration diagram showing the configuration of an English character recognition device in an embodiment of the present invention.

[Figure 2] Overall flowchart showing the control procedure of this embodiment

[
FIG. 3: Flowchart showing the control procedure for character area extraction processing in this embodiment

[Explanation of symbols]

1 Image input section 2 Document area extraction section 3 Word area extraction part 4 Character area extraction part 5 Character recognition section 6 Control section 7. Word dictionary 8 Character width dictionary 9 Recognition dictionary 10 Output section

Claims (1)

[Claims] Claims 1: An image input unit that inputs a document to be recognized; a text area extraction unit that outputs a document area from the input document image; a word area extraction unit that outputs a word area from the text area; A character area extraction unit that outputs a character area based on an English word dictionary that stores the spellings of English words and a character width dictionary that stores the character width of each English letter, and an image of the extracted character area. a character recognition unit that performs recognition based on a recognition dictionary that stores all characters from the character recognition unit; a control unit that controls the character recognition unit and the character area extraction unit to determine a recognized character code; and a control unit that determines a recognized character code. An English character recognition device consisting of an output section that outputs a code.