JP2982221B2 - Character reader - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information input device used for a word processor or an electronic typesetting system, particularly to an optical character reading device.

2. Description of the Related Art In recent years, word processors and electronic typesetting systems have become widespread, and the burden of input due to an increase in the amount of information has been called out, and there has been a demand for more efficient input. In addition, there is an increasing need to support document processing by a computer in order to effectively use document information and to efficiently store and search documents. In order to meet the demand, an optical character reading device capable of inputting a printed document independent of a document format, a document typeface, and the like is required. 2. Description of the Related Art In a conventional optical character reading apparatus, a method of extracting characters based on black pixel connected component information and a method of extracting characters from an average of character widths for each character are used in character extraction.

The operation of the conventional character segmentation method will be described below with reference to FIG. In the character segmentation method based on the black pixel connected component information, black pixels that are in contact with the black pixel 501 in eight directions (FIG. 5A) are distinguished based on the rule that they are the same single character pattern. For example, as shown in FIG.
Indicates one character pattern, and 503 indicates one character pattern. As described above, the conventional method is a method of extracting characters by assuming that the average of the widths of the circumscribed rectangular frames obtained from the black pixel connected components is the width of one character.

However, in the above-described conventional method, however, the 8
Since the black pixels in the directions are regarded as the same one character pattern,
The touching character is regarded as one character. For this reason, there has been a problem that characters cannot be contacted with each other and a correct character cannot be cut out from a complicated character.

SUMMARY OF THE INVENTION An object of the present invention is to provide a character reading device capable of performing accurate character segmentation with a simple configuration while keeping the above problems in mind.

Means for Solving the Problems In order to achieve the above object of the present invention, a shape detection unit that determines the shape of the left side of a character pattern, a character estimation unit that estimates a character from the shape, A character pattern guessed by a character guessing unit comprising a character cutout position detecting unit for obtaining a cutout position is cut out at a position determined by a character cutout position determining unit, and character recognition is performed.
It is a character reading device.

In the character reading device of the present invention having the above-described configuration, the shape detection unit scans the character pattern to be recognized from left to right with the sensor, sequentially moves the character pattern downward, and scans one surface. The feature is detected by the scanning distance from left to right until the detection, and a character pattern is estimated from dictionary data stored in the character estimation unit based on the detected left side shape. The character extraction position determination unit determines the extraction position from the inferred character pattern and recognizes it separately from other characters.
This makes it possible to obtain a correct character cut-out position even when the character pattern touches or the character pattern has a complicated outer shape.

Embodiment FIG. 1 is a block diagram showing a configuration of a character reading device according to an embodiment of the present invention. As shown in FIG. 1, reference numeral 100 denotes a character pattern input unit, which inputs a pattern for performing a character cutout process. Reference numeral 101 denotes a circumscribed rectangle detection unit that obtains a rectangular frame circumscribing the character pattern input from the character pattern input unit 100, 102 denotes a number-of-characters-in-rectangle detection unit that detects the number of characters in the circumscribed rectangle, and 103 denotes a left side of the input character pattern. A shape detection unit 104 for detecting a surface shape, a character estimation unit 104 for estimating a character from the shape obtained by the shape detection unit 103, a character extraction position determination unit 105 for obtaining an extraction position from the character estimated by the character estimation unit 104, 106 Is a display unit for displaying the cutout position of characters, and 107 is a dictionary for performing character estimation.

Hereinafter, the operation of this embodiment will be described using the block diagram of FIG. 1 and the cut-out example character pattern “fi” of FIG. 4A.

First, the binary pattern “fi” is input to the character pattern input unit 100.
Is entered.

Next, the circumscribed rectangle detection unit 101 performs a labeling process on the binary pattern, detects a pixel having the same label, and detects a rectangle circumscribed to the pixel. FIG. 4A shows a circumscribed rectangle of the character pattern "fi".

The number-of-characters-in-rectangle detection unit 102 of the circumscribed rectangle detects whether or not the characters in the rectangle are one character, that is, whether the characters are not in contact with each other or whether the characters are not complicated. As a detection method, when the input character pattern width is smaller than twice the average of the widths of the circumscribed rectangles of other character patterns, it is regarded as one character.
Assume at least characters. The following processing is performed on an input character pattern regarded as two or more characters, and a character regarded as one character is not processed by the shape detection unit 103, the character estimation unit 104, and the character cutout position determination unit 105. Move the processing to 106.

The shape detection unit 103 scans from left to right starting from the upper left end of the left side of the circumscribed rectangle, and has a primary characteristic as a distance to a position where the character pattern first meets the character pattern (FIG. 4B). . If it does not collide with the character pattern, the width of the circumscribed rectangle is taken as the primary feature. Similarly, scanning is performed from the left side of the circumscribed rectangle, and the distance to the second collision with the character pattern is used as a secondary feature (FIG. 4C).
If it does not collide with the character pattern, the width of the circumscribed rectangle is taken as the secondary feature. This is repeated up to the bottom of the circumscribed rectangle.

The dictionary 107 stores the primary features, the secondary features, and the corresponding aspect ratios. The aspect ratio is a value previously obtained from the aspect ratio of the circumscribed rectangle of the standard character pattern.

Next, the character estimating unit 104 compares the characteristic amount obtained from the input character pattern with the dictionary 107 created in advance from the standard pattern, and searches for a character having the closest characteristic amount. In the case of the input character pattern "fi", the character "1" is obtained.

The character cutout position determining unit 105 multiplies the horizontal / vertical ratio of the standard pattern of the character obtained by the character estimating unit by the height of the input character pattern (points 400 to 401 in FIG. 4D). A value is calculated, and a value on the right side of the input pattern from the left end by the value is set as a character cutout position (from point 402 to point 403 in FIG. 4D). In the case of the input character pattern "fi", a value obtained by multiplying the height / width of the input character pattern "fi" by the width / aspect ratio of the standard character pattern of "f" is calculated by guessing the character "f", The position is on the right side by the value from the left end of the input pattern. From this position to the right side of the circumscribed rectangle is the character pattern of the subsequent character "i".

Next, the processing flow of the present embodiment will be described in detail based on the flowcharts shown in FIGS. In step S200, a binarized character pattern is input. In step S201, a rectangular frame circumscribing a character pattern is determined by a black pixel connected component. Step S202
In, the number of characters in the circumscribed rectangle is detected. In step S203, it is determined whether the number of characters in the circumscribed rectangle is one. If it is correct, the process proceeds to step S205; otherwise, the process proceeds to step S204. Step S204 is a character separation subroutine. In step S205, it is determined whether the process has been completed for all circumscribed rectangles. If the process has been completed, the process proceeds to step S206; otherwise, the process proceeds to step S202.
Move the process to Step S206 ends.

The character separation subroutine is shown in the flowchart of FIG. 3 and will be described below. In step S301, a character pattern determined to be two or more characters is input. In step S302, primary and secondary features of the horizontal character pattern are obtained. Next, in step S303, characters are estimated. In step S304, the character cutout position is determined by determining the character guess.

As described above, in the present embodiment, the cutout position is determined after estimating the character, so that accurate cutout is possible.

Advantageous Effects of the Invention The present invention provides a character detecting unit and a character estimating unit that detect the shape of a left side surface. First, character estimation of an input character pattern is performed. A character reading apparatus characterized in that correct character segmentation can be performed even for a character pattern, and that the effect of improving the recognition rate of optical character reading can be obtained by estimating characters.

[Brief description of the drawings]

FIG. 1 is a block diagram showing the configuration of a character reading apparatus according to an embodiment of the present invention, FIGS. 2 and 3 are flowcharts of the embodiment, and FIG. 4 is a diagram until a cutout position is determined by the embodiment. FIG. 5 is a character pattern diagram for explaining a conventional example. 100: character pattern input unit, 101: circumscribed rectangle detection unit
102: Detecting unit for the number of characters in a rectangle 103: Detecting unit 104
... Character guessing section, 105... Character cutout position determining section.

Claims (3)

(57) [Claims] A shape detection unit configured to detect a left side shape of the recognition target character pattern; a character estimation unit configured to estimate a character pattern from a left side shape of the recognition target character pattern detected by the shape detection unit; A character cutout position determining unit that determines a character cutout position from the character pattern guessed by the character guessing unit, and performs character extraction by separating each character at the cutout position determined by the character cutout position determining unit. Character reading device. 2. The character reading device according to claim 1, wherein the character cutout position determining unit determines the character cutout position based on the horizontal / aspect ratio of the standard pattern of the character pattern estimated by the character estimating unit. 3. The character reading device according to claim 1, wherein the character estimating unit has a dictionary in which a standard character pattern including data of a left side shape of the character pattern and data of an aspect ratio is stored.