JPH06348911A - English character recognition device - Google Patents

Abstract

PURPOSE:To provide the English character recognition device which can be improved in recognition precision and shortened in processing time by automatically deciding the style of an document to be recognized. CONSTITUTION:This device is equipped with a circumscribed rectangle detection part 3 which detects circumscribed rectangles in binary image data, a line rectangle segmentation part 4 which segments a line rectangle from the distribution of the circumscribed rectangles, a character rectangle segmentation part 5 which segments a character rectangle from the line rectangle, etc., by integrating the circumscribed rectangles, a figure feature extraction part 6 which extracts figure features from the character rectangle, a character recognition part 8 which recognizes a character by comparing the figure features with a recognition dictionary 7 and stores the result in a recognition result storage part 9, a document style decision part 14 which decides the style of the object document from the distance from the center line of an adjacent character rectangle, a word detection part 15 which detects a word from this decision result and the interval between the character rectangles, and a misrecognized character correction part 12 which compares the detected word with a word dictionary 11 and corrects the recognition result in the recognition result storage part 9.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an English character recognizing device capable of recognizing typed characters such as newspapers and novels and English characters such as dot characters.

[0002]

2. Description of the Related Art In recent years, with the advent of an internationalized society, there is an increasing number of opportunities to refer to documents written in alphabets such as English and German (hereinafter referred to as English letters) to improve workability. Documents are being digitized. As an input device to the computer at this time, an English character recognition device has come to be used instead of the keyboard.

A conventional English character recognition device will be described below. FIG. 6 is a functional block diagram of a conventional English character recognition device. Reference numeral 1 denotes an image input unit that inputs a recognition target document as an image and outputs binary image data, 2 denotes an image storage unit that stores the binary image data output from the image input unit 1, and 3 denotes an image storage unit 2. A circumscribing rectangle detector 4 for detecting a circumscribing rectangle circumscribing a series of black pixels from the stored binary image data is a portion having a high horizontal position distribution of the circumscribing rectangle detected by the circumscribing rectangle detector 3. A line rectangle cutout unit that detects a line rectangle that is detected as a line and circumscribes each line is integrated based on the line rectangle cut out by the line rectangle cutout unit 4 and the circumscribed rectangles detected by the circumscribed rectangle detection unit 3 are integrated into a binary value. A character rectangle cutout unit that cuts out a character rectangle circumscribing one character in the image data, 6 is a graphic feature extraction unit that extracts the distribution of black pixels in the character rectangle cut out by the character rectangle cutout unit 5 as a graphic feature, 7 Is all The recognition dictionary 8 stores the graphic features of the recognition target character. Reference numeral 8 compares the graphic feature of the character rectangle extracted by the graphic feature extraction unit 6 with the graphic feature of the recognition target character in the recognition dictionary 7 to find the highest similarity. A character recognition unit that outputs a character code such as an ASCII code corresponding to a high recognition target character as a recognition result, a recognition result storage unit 9 that stores the recognition result output from the character recognition unit 8, and a character rectangle cutout unit 5 A word detection unit that detects individual words in the binary image data based on the horizontal intervals of the character rectangles cut out in step 11, a word dictionary that stores the correct spelling of all words, and 12 word detection It is determined by referring to the word dictionary 11 whether or not the spelling of the word detected in the section 10 is correct, and it is determined that the misspelled word includes the misrecognized character, and the recognition result is obtained. Misrecognized character correction unit corrects the characters in 憶部 9, 13 is a recognition result output unit for outputting the recognition result stored in the misrecognized character correction unit 12 the recognition result storage section 9, which is corrected by.

The operation of the conventional English character recognizing device constructed as above will be described below. First, the image input unit 1 converts a recognition target document into a binary image, inputs the binary image, and stores the binary image in the image storage unit 2. Next, the circumscribing rectangle detecting unit 3 detects a circumscribing rectangle circumscribing a series of black pixels in the binary image data stored in the image storage unit 2. Next, the line rectangle cutout unit 4 obtains the distribution of the horizontal position of the circumscribed rectangle detected by the circumscribed rectangle detection unit 3, cuts out a portion having a high distribution as a line, and detects a line rectangle circumscribing each line. Next, the circumscribed rectangle detected by the circumscribed rectangle detection unit 3 is determined by the character rectangle cutout unit 5 to determine the size of the circumscribed rectangle, the distance between the circumscribed rectangles, and the line rectangle cutout unit 4.
The character rectangles circumscribing one character in the binary image data are extracted by integrating them based on the line rectangles and the like extracted by. next,
The figure feature extraction unit 6 extracts the distribution of black pixels of the character image in the character rectangle cut out by the character rectangle cutout unit 5 as a figure feature. Next, the character recognition unit 8 compares the graphic feature of the character image in the character rectangle extracted by the graphic feature extraction unit 6 with the graphic feature of the recognition target character stored in the recognition dictionary 7 to determine the similarity. The character code such as the ASCII code corresponding to the highest character to be recognized is output as a recognition result and stored in the recognition result storage unit 9. Next, the word detection unit 10 detects each word, assuming that there is a word break in a portion where the horizontal interval of the character rectangle cut out by the character rectangle cutout unit 5 is larger than the word break threshold value. . Next, the misrecognized character correction unit 12 checks whether or not the spelling of each word detected by the word detection unit 10 is correct by referring to the word dictionary 11, and among the words determined to be misspelled, If the recognition accuracy of the character that seems to have been erroneously recognized is low, the character is corrected to a correct character estimated from the word dictionary 11. Next, the recognition result output unit 13 causes the erroneously recognized character correction unit 1
The recognition result stored in the recognition result storage unit 9 corrected in 2 is output.

[0005]

However, in the above-described conventional configuration, there are two types of styles of English documents to be recognized, that is, mono-space and proportional, and there are mono-space and proportional documents. Because the intervals of the circumscribing rectangles are different, the word detection unit may make a false detection when detecting a break between words.
There is a problem that the recognition accuracy is lowered and the reliability is low, and the processing time for word detection is long and the workability is low.

The present invention solves the above-mentioned problems of the prior art by automatically determining whether the style of the document to be recognized is mono-space or proportional to improve the recognition accuracy and reliability. An object of the present invention is to provide an English character recognizing device which is excellent in workability and which can reduce processing time.

[0007]

In order to achieve this object, the English character recognition apparatus of the present invention stores an image storage unit for storing binary image data obtained from a document to be recognized, and the image storage unit. The circumscribing rectangle detecting unit that detects a circumscribing rectangle that circumscribes a series of black pixels in the binary image data, and a portion in which the horizontal distribution of the circumscribing rectangles detected by the circumscribing rectangle detecting unit is high is defined as a row A line rectangle cutout unit that detects a line rectangle that circumscribes each line, and a circumscribed rectangle that is detected by the circumscribed rectangle detection unit based on the line rectangle that is cut out by the line rectangle cutout unit is integrated, A character rectangle cutout part that cuts out a character rectangle that circumscribes two characters,
A graphic feature extraction unit that extracts the distribution of black pixels in the character rectangle cut out by the character rectangle cutout unit as a graphic feature, a recognition dictionary that stores the graphic features of all recognition target characters, and the graphic feature extraction. A character recognition unit that performs character recognition by comparing the graphic features of the character rectangle extracted by the character recognition unit with the graphic features of the recognition target character in the recognition dictionary; and a recognition result storage that stores the recognition result of the character recognition unit. And a document style determination unit that determines whether the style of the recognition-target document is mono space or proportional from the distance between the center and the center line of the adjacent character rectangle of the character rectangle cut out by the character rectangle cutout unit. A word detection unit that detects individual words based on the determination result of the document style determination unit and the horizontal spacing of the character rectangles cut out by the character rectangle cutout unit, and the like. A word dictionary that stores the correct spelling of all words and a character that is erroneously recognized by comparing the word detected by the word detection unit with the word dictionary are detected, and the recognition result in the recognition result storage unit And an erroneously recognized character correction unit for correcting

[0008]

With this configuration, the document style determination unit automatically determines the style of the recognition target document based on the distance between the center lines of the adjacent character rectangles of the character rectangles cut out by the character rectangle cutout unit. Since the word detection unit detects individual words according to the determined style, the words can be detected accurately, the recognition accuracy can be improved, and the processing time can be shortened. .

[0009]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An English character recognition apparatus according to an embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a functional block diagram of an English character recognition device according to an embodiment of the present invention, and FIG. 2 is a device block diagram of an English character recognition device according to an embodiment of the present invention. 1 is an image input unit, 2 is an image storage unit, 3 is a circumscribed rectangle detection unit, 4 is a line rectangle cutout unit,
5 is a character rectangle cutout unit, 6 is a graphic feature extraction unit, 7 is a recognition dictionary, 8 is a character recognition unit, 9 is a recognition result storage unit, 11 is a word dictionary, 12 is a misrecognized character correction unit, and 13 is a recognition result output. Since these are the same as those in the conventional example, the same reference numerals are given and description thereof is omitted. Reference numeral 14 is a document style for determining whether the style of the recognition target document is mono space or proportional based on the interval between the center lines of the adjacent character rectangles of the character rectangles cut out by the character rectangle cutout unit 5. Judgment unit, 15 is a document style judgment unit 14
In the case of mono-spaced documents, the part where the distance between the center lines of adjacent character rectangles is large is defined as the word break in the case of mono space documents, and in the case of proportional documents, the part where the space between adjacent character rectangles is large is defined as the word break. It is a word detection unit that detects individual words as word breaks. In FIG. 2, 16
Is a central processing unit (hereinafter abbreviated as CPU) that controls the entire English character recognition device, and 17 is a control program region 18 in which a program in which instructions to the CPU 16 are written and a recognition dictionary region in which the recognition dictionary 7 is stored 19 and word dictionary 1
A read-only memory (hereinafter abbreviated as ROM) having a word dictionary area 20 in which 1 is stored, 21 is a binary image data area 22 and a line rectangle cutout section 4 which constitute the image storage section 2.
A line rectangle data area 23 in which the data of the line rectangle cut out by the above is stored, and a character rectangle data area 24 in which the data of the character rectangle cut out by the character rectangle cutout unit 5 is stored,
A random access memory (hereinafter abbreviated as RAM) 27 having a word data area 25 in which data of individual words detected by the word detection unit 15 is stored, and a recognition result data area 26 constituting the recognition result storage unit 9; Is a scanner that constitutes the image input unit 1 and scans a document to be recognized to obtain binary image data; 28 is a keyboard for giving an instruction to start character recognition to an English character recognition device; and 29 is a recognition result output unit 13
And an output device for outputting the recognition result, and a bus 30 for connecting the above devices.

The operation of the English character recognition apparatus according to the embodiment of the present invention constructed as above will be described below. FIG. 3 is an overall flowchart of the English character recognition apparatus according to the embodiment of the present invention, and FIG. 4 is a detailed flowchart of the document style determination processing of the English character recognition apparatus according to the embodiment of the present invention. 5B is a diagram showing an example of a proportional document, and FIG. 5B is a diagram showing an example of a mono space document. d1 and d2 are distances between center lines of character rectangles indicating distances between center lines of adjacent character rectangles.

First, the image input unit 1 converts a document to be recognized into a binary image and inputs it, and the image is stored in the image storage unit 2.
(S1). Next, the circumscribing rectangle detection unit 3 obtains a circumscribing rectangle circumscribing a series of black pixels in the binary image data stored in the image storage unit 2 (S2).
Next, the line rectangle cutout unit 4 obtains the distribution of the horizontal position of the circumscribed rectangle detected in S2, cuts out a portion having a high distribution as a line, detects the line rectangle circumscribing each line, and outputs the data. Store in the rectangular data area 23 (S
3). Next, the character rectangle cutout unit 5 integrates the circumscribed rectangles based on the size and distance of the line rectangle detected in S3 and the circumscribed rectangle detected in S2, and circumscribes one character in the binary image data. The character rectangle is cut out and stored in the character rectangle data area 24 (S4). Next, the graphic feature extraction unit 6 extracts a character image for one character from the binary image data in the image storage unit 2 based on the position coordinates of the character rectangle stored in the character rectangle data area 24. Then, the distribution of the black pixels of this character image is obtained to extract the graphic feature (S5). Next, the character recognition unit 8 compares the graphic feature of the character image extracted in S5 with the graphic feature of the recognition target character in the recognition dictionary 7 in the same manner as in the conventional example, and the recognition target having the highest degree of similarity is compared. A character code such as an ASCII code corresponding to the character is output as a recognition result and stored in the recognition result storage unit 9 (S6). Next, the document style determination unit 14 determines whether the style of the recognition target document is mono space or proportional based on the positional relationship of the character rectangles cut out in S4 (S
7). This Step will be described later. Next, the word detection unit 15 determines that the style determined in S7 is mono.
If it is a space, the part where the distance between the center lines of adjacent character rectangles is large is detected, and if it is proportional, the part where the space between adjacent character rectangles is large is detected as the word delimiters. , Are stored in the word data area 25 (S8). Next, the misrecognized character correction unit 12 checks whether or not the spelling of the word detected in S8 is correct by referring to the word dictionary 11, and corrects the spelling error in the word determined to be incorrect. When the recognition accuracy of the character that seems to be recognized is low in S6, this character is corrected to the correct character estimated from the word dictionary 11 (S9). Next, the recognition result output unit 13 outputs the recognition result corrected in S9 and stored in the recognition result storage unit 9 from the output device 29, and all processing is terminated (S10).

Next, the document style determination process in S7 will be described in detail. In FIG. 4, first, the array variable Table [] is initialized as a frequency distribution table (S11). Next, the variable Spcno is cleared to zero as a counter for the number of spaces between characters (S12). Next, the distance between the center lines of the adjacent character rectangles cut out in S4 is substituted into the variable d (S13). Here, as an example, d1 is d1 between'c 'and'i' in a proportional recognition target document as shown in FIG. 5A, and a mono space as shown in FIG. Similarly in the document to be recognized, d
And d2 are respectively substituted into. Next, d substituted in S13 is divided by 4, and the value rounded off is substituted for the variable d '(S14). Next, the d'th (hereinafter referred to as Table [d ']) of the frequency distribution table is incremented by 1 (S15). Next, the inter-character space number counter Spcno is incremented by 1 (S1
6). Next, it is checked whether the processing for all the character rectangles has been completed (S17). If No, jump to S13 to perform processing for the next character rectangle, and if Yes, the frequency Table [x] in the frequency distribution table.
Find the position x where the value of is the maximum, and use this as the variable maxno
(S18). Next, set the variable max to Table
[Maxno] That is, the maximum value of the frequency is substituted (S1
9). Next, it is checked whether the value of maxno is less than or equal to the contact character consideration threshold value (S20). If No, S2
If it is Yes, then it is checked whether the value obtained by dividing max by the intercharacter space counter Spcno is larger than the style determination threshold value (S21). If No, jump to S25, and if Yes,
Judge that the style of the recognition target document is mono space,
The document style determination process ends (S22). Next, the value of Table [maxno × 2] is added to max in consideration of the possibility that adjacent characters in the recognition-target document come into contact with each other (S23). Next, it is checked whether the value obtained by dividing max by the inter-character space counter Spcno is larger than the contact character considered style determination threshold value (S24). If Yes, jump to S22, and if No,
The style of the recognition target document is determined to be proportional, and the document style determination process ends (S25).

[0013]

As described above, according to the present invention, the style of the document to be recognized is determined based on the distance between the center lines of the adjacent character rectangles among the character rectangles cut out by the character rectangle cutout unit by the document style determination unit. Is automatically determined, and the word detection unit detects individual words according to the determined style, so that it is possible to perform accurate word detection, improve recognition accuracy, and have excellent reliability. At the same time, it is possible to realize an English character recognizing device with excellent workability that can reduce the processing time.

[Brief description of drawings]

FIG. 1 is a functional block diagram of an English character recognition device according to an embodiment of the present invention.

FIG. 2 is a device block diagram of an English character recognition device according to an embodiment of the present invention.

FIG. 3 is an overall flowchart of an English character recognition device according to an embodiment of the present invention.

FIG. 4 is a detailed flowchart of document style determination processing of the English character recognition device according to the embodiment of the present invention.

FIG. 5A is a diagram showing an example of a proportional document. FIG. 5B is a diagram showing an example of a mono space document.

FIG. 6 is a functional block diagram of a conventional English character recognition device.

[Explanation of symbols]

 1 image input unit 2 image storage unit 3 circumscribed rectangle detection unit 4 line rectangle cutout unit 5 character rectangle cutout unit 6 figure feature extraction unit 7 recognition dictionary 8 character recognition unit 9 recognition result storage unit 10 word detection unit 11 word dictionary 12 misrecognition Character correction unit 13 Recognition result output unit 14 Document style determination unit 15 Word detection unit 16 Central processing unit (CPU) 17 Read only memory (ROM) 18 Control program area 19 Recognition dictionary area 20 Word dictionary area 21 Random access memory (RAM) ) 22 binary image data area 23 line rectangular data area 24 character rectangular data area 25 word data area 26 recognition result data area 27 scanner 28 keyboard 29 output device 30 bus

Claims (1)

[Claims] 1. An image storage unit for storing binary image data obtained from a document to be recognized, and a circumscribing rectangle circumscribing a series of black pixels in the binary image data stored in the image storage unit are detected. A circumscribed rectangle detection unit, a line rectangle cutout unit that detects a portion having a high horizontal position distribution of the circumscribed rectangle detected by the circumscribed rectangle detection unit as a line, and cuts out a line rectangle that circumscribes each line, and the line rectangle cutout Based on the line rectangle cut out by the unit, the circumscribed rectangles detected by the circumscribed rectangle detection unit are integrated to cut out a character rectangle circumscribing one character, and a character rectangle cutout unit is cut out by the character rectangle cutout unit. A graphic feature extraction unit that extracts the distribution of black pixels in a character rectangle as a graphic feature, a recognition dictionary that stores the graphic features of all recognition target characters, and a sentence extracted by the graphic feature extraction unit. A character recognition unit that performs character recognition by comparing a rectangular graphic feature with a graphic feature of a recognition target character in the recognition dictionary, a recognition result storage unit that stores a recognition result in the character recognition unit, and the character rectangle. A document style determination unit that determines whether the style of the recognition target document is mono space or proportional from the distance between the center lines of the adjacent character rectangles of the character rectangles cut out by the cutout unit, and the document style determination unit. A word detection unit that detects individual words based on the determination result in (1) and the horizontal intervals of the character rectangles cut out by the character rectangle cutting unit, and a word dictionary that stores the correct spelling of all words. And a erroneously recognized character correction unit that corrects the recognition result in the recognition result storage unit by comparing the word detected by the word detection unit with the word dictionary to detect erroneously recognized characters. English character recognition apparatus characterized by comprising a.