JPH05282493A - Roman letter recognizing device - Google Patents

Abstract

PURPOSE:To provide the Roman letter recognizing device which improves the operability of letter input by recognizing a document of low quality which contains contacting letters and blurring letters at a high speed. CONSTITUTION:This Roman letter recognizing device consists of a means which segments a row area after segmenting a document area, and then estimates the letter width of the letter with the largest letter width according to the height of the row, a means which segments a letter area almost to the estimated letter width and recognizes the area, a means 11 which stores the height of the letter area based upon a reference letter, the position (letter position feature) based upon a reference line, and image information when the recognition result is acceptable, and a means 10 which limits an object letters to be segment again in the letter area according to a letter information dictionary 9 containing letter position features of all letters, superposes the stored letter image and segments the letter again, and performs recognition when the recognition result is rejected.

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 for recognizing characters in English general documents.

[0002]

2. Description of the Related Art In recent years, there has been an increasing demand for using a character recognition device as an input device for a computer or the like, and a character input device capable of efficiently obtaining a stable recognition result improves the performance of a system such as a computer. Has become indispensable to.

A conventional English character recognizing device reads from a document area
When cutting out a character area, paying attention to the fact that English characters are not separated into left and right, the area surrounded by white spaces is cut out as a character area and recognition processing is performed.

[0004]

However, in the above-mentioned conventional configuration, two characters on the left and right of the document are in contact with each other (hereinafter, these characters are referred to as contact characters) in a document whose print quality is deteriorated due to copying or the like. One character may be cut and divided into two (hereinafter, these characters are referred to as cut characters). In this case, the conventional method cannot correctly cut out the character area. ..

Further, since the recognition accuracy is low, the recognition cannot be performed at high speed, and the workability of character input is low.

The present invention solves the above-mentioned conventional problems, and in character recognition of an English general document, character input capable of high-speed recognition of a low-quality document including contact characters and broken characters. It is an object of the present invention to provide an English character recognizing device whose workability is remarkably improved.

[0007]

In order to achieve this object, the English character recognizing device of the present invention cuts out a document area, then cuts out a line area, and has the widest character width based on the height of the line. Guess the character width. Then, the character area is cut out before and after the estimated character width, and recognition is performed. If the recognition result is worthy of the recognition result, the height of this character area with respect to the reference character, the position with respect to the reference line (hereinafter, these height and position information are referred to as character position features), and the image information are stored. .. If rejected (rejected), the characters to be recut are limited based on the character information dictionary that stores the character position characteristics of all characters in advance for this character area, and the stored character image By overlapping and re-extracting and performing recognition processing, the accuracy of extracting a character area of a document including contact characters and cut characters and the processing speed are improved. Specifically, an image input unit for inputting a document to be recognized, a text region cutout unit for outputting a text region from the input document image, a line region cutout unit for outputting a line region from the text region, and a line region A character area cutout unit that outputs a character area, a recognition dictionary that stores graphic features of all recognition target characters in advance, and a character recognition unit that performs recognition from an image of the cut out character area based on the recognition dictionary. And a re-cutout area storage section that stores the position of the area where the character area needs to be cut out and the character recognition is performed again as a result of recognition in the character recognition section, and the character recognition section determines that the recognition result is worth it. A character information storage unit that stores the character recognition character code, the height of the character with respect to the reference character, the position of the character with respect to the reference line (hereinafter referred to as the character position feature), and the character image. A character information dictionary in which the character position characteristics of the recognition target character are stored in advance, and the recut target character is limited to the area stored in the recut area storage unit by using the character information dictionary, and the character An overlay processing unit that overlays the images stored in the information storage unit to perform clipping and recognition,
The recognition result storage unit stores the result and position of character recognition, and the recognition result output unit outputs the recognition result stored in the recognition result storage unit.

[0008]

With this configuration, when the character region is cut out, the maximum character width is estimated and the character region is cut out, so that the cut character can be cut out without being divided, and the recognition result is rejected. In this case, the character position feature is used for this character area to limit the characters to be re-cut out, and the images of highly recognizable characters in the same document are overlaid, and the cut-out character and contact character are cut out at high speed. ,
Can be recognized.

[0009]

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

FIG. 1 is an overall configuration diagram of an English character recognition apparatus of this embodiment. 1 is an image input unit for inputting a document to be recognized as a document image, 2 is a sentence region cutout unit for finding a set of character strings from the input document image, and outputting a sentence region, 3 is a line unit break from the sentence region A line area cutout unit that finds and outputs the range of one line as a line area, 4 is a character area cutout unit that finds a character-based delimiter from the line area, and outputs a range of one character as a character area, and 5 is all recognition targets A recognition dictionary 6 in which graphic features of characters are stored in advance, 6 is used to compare the graphic features in the character region cut out by the character region cutting unit 4 with the graphic features in the recognition dictionary 5 to determine the similarity between them. A character recognition unit that seeks and recognizes
Reference numeral 7 is a re-cutout area storage unit that stores the position of the area where the character area needs to be cut out again as a result of recognition by the character recognition unit 6, and 8 is a recognized character code of a character with high recognition accuracy as a result of character recognition. , A character height with respect to a reference character, a character position with respect to a reference line (hereinafter, these character heights and character positions are referred to as character position features), and a character information storage unit for storing images, 9 are all The character information dictionary 10 in which the character position features of the recognition target characters are stored in advance for the area stored in the re-cut area storage unit 7,
Based on the character information dictionary 9, the characters to be re-cut out are limited, the image information stored in the character information storage unit 8 is overlapped, a character region is cut out, and recognition is performed. And a recognition result storage unit that stores the position, a recognition result output unit 12 that stores the recognition result based on the recognition result storage unit, and an internal bus 13 that connects the units 1 to 5, 7 to 9 and 11 , 14, 15, 16 is 5
It is an internal bus that connects 6 and 9 and 10 and 11 and 12.

The operation of the English character recognizing device configured as described above will be described below with reference to the drawings.

FIG. 2 is an overall flow chart, and FIG.
FIG. 4 is a flowchart for character area cutout / recognition processing, FIG. 4 is a flowchart for character area recutout / recognition processing, FIG. 5 is a flowchart for overlay processing, and FIG. 6 is an explanatory diagram of character position features. FIG. 7 is an explanatory diagram showing the contents of the character information dictionary.

First, in FIG. 2, a document to be recognized is input as a document image to the image input section 1 (S1).

The input document image is sent to the text area cutout unit 2, and the text area cutout unit 2 obtains a histogram of black pixels in the vertical and horizontal directions of the document image, and the text area is extracted based on this histogram. .. Then, the position information of the text area is stored as internal data (S2).

The position information of the text area is sent to the line area cutout unit 3, and the line area is cut out from the text area. The line area cutout unit 3 obtains a horizontal histogram in the text area, and cuts out the line area using a minimum value close to 0 as a line break. The position information of all line areas in the text area found by the text area cutout unit 2 is obtained and stored as internal data (S3).

Next, the character area cutout processing / character recognition processing (S4) will be described with reference to the flowchart of the character area cutout and recognition processing of FIG.

Position information of the line area is sent to the character area cutout unit 4 to cut out the character area. The character area slicing unit 4 estimates the maximum character value as follows based on the position information of the sent line area (S9).

Maximum character width = line height × 1.2 The circumscribed rectangle of the black pixel connected portion of the image data in the line area is extracted (S10). If there is a small rectangle in the upper 1/3 or lower 1/3 of the line and there is a rectangle in the vertical direction,
The circumscribed rectangles are integrated by regarding them as separated characters such as i and j (S11). Hereinafter, this integrated rectangle is referred to as a basic rectangle. If the basic rectangle to be processed is at the beginning of a line, the X coordinate of the left side of the leftmost rectangle in the line region is the cutout position. If it is not at the beginning of the line, the left side of the circumscribed rectangle to be processed next is The X coordinate of the side is the cutout position. The basic rectangles on the right side of this cutout position are integrated in order from the rectangle on the left within a range that does not exceed the maximum character width. The integrated area is cut out as a character area and the position information is stored as internal data. If only one circumscribed rectangle is larger than the maximum character width, the one circumscribed rectangle is cut out as a character area (S12).

The position information of the character area is sent to the character recognizing unit 5 to perform the recognition process. The character recognition unit 6 extracts the graphic feature of the image in the character region based on the sent position information of the character region, and extracts the extracted graphic feature and the graphic feature of all the characters in the recognition dictionary 5. The similarity between the two is obtained (S13), and it is determined whether or not the recognition result is worth (S14). If the recognition result is worthy, it is checked whether or not the character position feature and the image of the character to be recognized are already stored in the character information storage unit 8 (S15). If the information of the target character area is not stored in the character information storage unit 8, the recognized character code, the character position feature and the image are stored in the character information storage unit 8 (S16). In S15, the recognized character code and the character position information are sent to the recognition result storage unit 11 regardless of whether or not the information of the target character is stored in the character information storage unit 8 (S17). In S14, if the recognition result is not merited, it is checked whether the character area immediately before is a re-cutout area (S18). If the immediately preceding area is the re-cutout area, this area and the area currently targeted are combined into one area (S19). The position information of the target character area (the area when integrated) is stored in the recut area storage unit 7 (S20).

The process of S4 is performed for all the characters in the text area (S5). Next, regarding the character area recutting / recognition processing (S6), the character area recutting / recognition processing of FIG.
A description will be given according to the flowchart of the recognition process and the flowchart of the superposition process of FIG.

The character position information of the recut character area stored in the recut area is sent to the character area slicing section 4 (S21). The character area cutout unit 4 stores the X coordinate on the left side of the received rectangle as the cutout position in the internal data (S22). The position information of the re-cutout area and the cutout position are sent to the superposition processing unit 10.

The character position feature and character information dictionary 9 used in the superposition processing will be described with reference to the character position feature explanatory diagram of FIG. 6 and the character information dictionary content diagram of FIG.

The baseline is a virtual line that serves as a reference,
"A", "b", "A", etc. indicate the line where the lower ends of most characters are located. The x-height indicates the height of the lowercase letter "x" (h2 in FIG. 6). The descender refers to the lower part of this baseline, and the ascender refers to the upper part of the x-height. The height of the capital letter (h1 in FIG. 6) is also used as the reference character of the height information. The character information dictionary 9 stores the height with respect to the baseline and the presence or absence of a descender for each character type to be recognized. The height information is divided into three, the height from the baseline from the height of uppercase letters and the height of lowercase letters, which cannot be determined, and is set to 1, 2, and 3, respectively. The descender information is 1, when there is no descender, Some cases are 2, and some are in typeface.

The superposition processing unit 10 checks whether the height of the re-cutout region that has been sent is equal to x height.
38). If the height of the re-cutout area is equal to x height, the characters included in this re-cutout area are regarded as x-height characters only, and the height of the character information dictionary of the images stored in the character information storage unit 8 is considered. Images of character types with a size of 1 or 3 and a descender of 1 or 3 are taken out in the order of increasing character width, and are superimposed in order from the cutout position (S3).
9). If the height of the re-cutout area is not equal to x height, the position where the descender or the ascender appears for the first time in the horizontal direction from the cutout position is found (S40).
If both the descender and the ascender do not exist (S4
1) The characters existing in this area are regarded as x-height characters, and superposition similar to S39 is performed. When the descender first appears from the cut-out position (S42), it is checked whether the horizontal distance from the cut-out position to the position where the descender first appears is less than or equal to the width of the lower case "y" (S43). If the horizontal distance from the cutout position to the position where the descender exists is less than or equal to the width of the uppercase “y”, it is considered that a character including the descender may exist, and the character information storage unit 8 first stores the character. Among the stored images, the descender of the character information dictionary 9 takes out images of character types 2 or 3 in the order of increasing character width, and superimposes them sequentially from the cutout position (S44). It is checked whether there is a matching image (S45). If there is no matching image, it is considered that there is an x-height character immediately behind the cutout position, and the same superimposition as in S39 is performed (S49). When it is the ascender that first appears from the cutout position, it is checked whether the horizontal distance from the cutout position to the position where the ascender appears is less than or equal to the width of the capital letter "W" (S46). If the horizontal distance from the cutout position to the position where the ascender exists is less than or equal to the width of the capital letter "W", it is considered that a character including the ascender may exist, and the character information storage unit 8 first stores the character. Among the stored images, images of character types having a height of 1 or 3 in the character information dictionary 9 are taken out in the order of increasing character width, and are sequentially superposed from the cutout position (S47). It is checked whether or not there is a matching image, but if there is no matching image, it is considered that there is an x-height character immediately behind the cut-out position, and superimposing similar to S39 is performed (S49).

The superimposing process is performed in this way, and it is checked whether or not there is a matching image (S24). If a matching image exists, the recognition result and the position information are sent to the recognition result storage unit 11 (S36). If there is no matching image, the number of basic rectangles existing in this re-cutout area is counted (S25). When the number of basic rectangles is 2 or more, the rightmost basic rectangle in this area is removed and the remaining basic rectangles are integrated. This integrated area is newly set as a character area, and position information is stored in internal data (S31). The position information of the new character area is sent to the character recognition unit 6 again, and the recognition processing similar to S31 is performed (S3
2) Determine whether the recognition result is worth it. When the recognition result is not merited, the processing from S25 is repeated (S3).
3). If the recognition result is worthy, it is checked whether or not the character position feature and the image of the character to be recognized are already stored (S34). When the character position feature and the character image of the target character are not stored in the character information storage unit 8, the recognized character code, the character position feature, and the image are stored in the character information storage unit 8 (S35). In S30, the recognition character code and the character position information are sent to the recognition result storage unit 11 regardless of whether the character position characteristics of the target character and the image are stored in the character information storage unit 8 (S36). When the number of basic rectangles is 1, for this re-cut area,
It is checked whether the re-cutout has been performed at least once (S26). If the re-cutout has been performed at least once, the recognition process similar to S13 is performed (S27), and it is determined whether or not the recognition result is worth (S28). If the recognition result is worth, the character position feature of the character to be recognized and the image are the same as above.
It is checked whether or not it is already stored (S34). If the character position feature and image of the target character are not stored in the character information storage unit 8, the recognized character code, character position feature and image are stored in the character information storage unit 8. Is stored (S35). S3
In 4, the recognized character code and the character position information are sent to the recognition result storage unit 11 regardless of whether or not the character position characteristic of the target character and the image are stored in the character information storage unit 8 (S36). In S26, if re-cutting has not been performed within this re-cutting area, or S
If the recognition result is not worth in 28, it is considered that a contact character exists in this area, and it is cut at the concave portion of the contour of the black pixel in the area to make a character area, and the position information is stored in the internal data (S29). .. The position information is sent to the character recognition unit 6 and the same recognition processing as S13 is performed (S30). If the recognition result is worth, the recognition character code and character position information,
When the recognition result is not worth, the rejection code and the position information are sent to the recognition result storage unit 11 and stored (S3).
6).

If the processed basic rectangle is not the rightmost rectangle among the basic rectangles included in the re-cutout area, the cutout position is on the left side of the cutout character area. The X coordinate of the side is set (S22), this position information is sent to the character area cutout unit 4, and the processing from S23 onward is repeated. Of the basic rectangles included in the processed re-cutout area, if the basic rectangle is located at the rightmost position, the re-cutout and recognition processing in this recutout area is terminated (S34).

The character region recutting / recognizing process (S6) is performed on all the recutting regions (S7).

The recognition result storage unit 11 sends the recognition character code and position information to the recognition result output unit 12. The recognition result output unit 12 outputs the recognized character code based on the sent position information (S8).

By performing the processing of S1 to S8,
Performs character recognition processing of the given document image.

[0030]

As described above, according to the present invention, when the character area of an English general document is cut out, the maximum character width is estimated and the character area is cut out so that the cut character is cut out without being divided. You can plan that. In addition, when the result of recognition is rejected, the character to be re-cut out is limited to this character area by using the character position feature, and it is superimposed with the image of the character with high recognition character accuracy in the same document. Thus, it is possible to realize an English character recognizing device which is capable of cutting out a broken character and a contact character at high speed and recognizing the character and excellent in workability of character input.

[Brief description of drawings]

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

FIG. 2 is an overall flowchart of the English character recognition device according to the present embodiment.

FIG. 3 is a flowchart of a character region cutout / recognition process step of the English character recognition device according to the present embodiment.

FIG. 4 is a flowchart of a character area re-cutting / recognition process step of the English character recognition apparatus according to the present embodiment.

FIG. 5 is a flowchart of a superposition processing process of the English character recognition device according to the present embodiment.

FIG. 6 is an explanatory diagram of character position features of the English character recognition device according to the present embodiment.

FIG. 7 is an explanatory diagram of contents of a character information dictionary of the English character recognition device according to the present embodiment.

[Explanation of symbols]

 1 image input section 2 text area cutout section 3 line area cutout section 4 character area cutout section 5 recognition dictionary 6 character recognition section 7 recutout area storage section 8 character information storage section 9 character information dictionary 10 overlay processing section 11 recognition result storage Part 12 Recognition result output part

Claims (1)

[Claims] 1. An image input unit for inputting a document to be recognized, a text region cutout unit for outputting a text region from the input document image, a line region cutout unit for outputting a line region from the text region, and a line region A character area cutout unit that outputs a character area, a recognition dictionary that stores graphic features of all recognition target characters in advance, and a character recognition unit that performs recognition from an image of the cut out character area based on the recognition dictionary. And a re-cutout area storage section that stores the position of the area where the character area needs to be cut out and the character recognition is performed again as a result of recognition in the character recognition section, and the character recognition section determines that the recognition result is worth it. Character recognition character code, height of character with respect to reference character, position of character with respect to reference line (hereinafter,
(Character position feature) and a character information storage unit that stores a character image, a character information dictionary that stores in advance character position features of all recognition target characters, and an area stored in the re-cutout area storage unit. For the re-cutting target character using the character information dictionary, a superposition processing unit for superposing and cutting out and recognizing images stored in the character information storage unit, and a result and position of character recognition. And a recognition result output unit for outputting the recognition result stored in the recognition result storage unit.