JPH04104387A - Character recognizing means - Google Patents

Abstract

PURPOSE:To recognize only the real character existing inside a frame by deciding character recognition area based on counted peripheral distribution. CONSTITUTION:The peripheral distribution of a character bit is counted, and frame distribution T1, T2, and t1, t2 and character area distribution Mm is detected by a decision reference value calculated by a prescribed expression. Next, when the character string area can be limited by deleting frame area T, t, each character is segmented from an original picture. Here, outline extraction upper/lower areas calculated by the prescribed expression are set, the character whose outline is extracted in the area can be judged as the real character, and the recognition of one character is terminated by the normalization of character and the matching with a pattern dictionary. Thus, accurate character recognition can be performed even when the frame component surrounding the character or vertical line noise exists.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to an apparatus for recognizing character images read by an optical sensor or the like as characters.

(Prior art) In conventional character recognition methods, if there is a frame surrounding a true character area in a read image, the frame is also recognized as a character when character is extracted, resulting in erroneous recognition. .

In addition, as a special method, which is especially adopted in form character recognition devices, characters to be recognized are written in a frame in advance, and the frame is illuminated with light of the same color as the original paper irradiation light for image reading. By using line components, the frame components have already been removed from the read character image.

(Problem to be Solved by the Invention) However, with this type of recognition means, when recognizing characters such as alphanumeric characters on a form written in a format that is not determined in advance, it may be misrecognized, and the format may be incorrect. The method of specifying lacks versatility, which causes an increase in costs.

Therefore, the present invention was made in view of such problems.
The purpose of this is to make it possible to recognize only the true characters that exist inside the frame without specifying the format of the frame, even for characters written on the original paper that has a frame, and if necessary. It is an object of the present invention to provide a method for recognizing the existence of a frame and the fact that a character area is surrounded by the frame.

(Means for Solving the Problems) The character recognition means of the present invention includes a character image reading means realized by an optical sensor or the like, a means for counting whether the input character area is a character in the sub-scanning direction, and the means for repeating the counting means in the main scanning direction; means for determining whether the main scanning direction component is vertical line noise or a frame surrounding a character based on the counted value in the main scanning direction; a means for determining an area to perform character recognition based on the peripheral distribution obtained by the character recognition; a means for extracting and normalizing the character by extracting the outline thereof; means for determining whether the image is a character based on upper and lower limit values for contour extraction in the image; a pattern dictionary or peripheral feature dictionary for character recognition; and means for comparing the normalized character with the recognition dictionary. and a character recognition means characterized by:

(Example) The details of the recognition means of the present invention will be explained below based on the illustrated example.

FIG. 1 is a flowchart of the character recognition means of the present invention. FIG. 2 shows an example of a character image and an example of the marginal distribution count results obtained by the image manual 101. FIG. 3 shows an example of a character image input diagonally and an example of marginal distribution counting results obtained by the image input 101. FIG. 4 shows an example of an image without a character frame and an example of the counting results of the peripheral distribution.

Character recognition starts by acquiring a character image through image input 101. Image input is acquired by a reading device realized by an optical sensor or the like. There are two types of reading devices: one that scans the paper surface automatically and one that scans the paper surface manually.If the reading device is manual, the image may be input diagonally as shown in the image in FIG. When an image is input diagonally, there is a method of applying a rotation correction method to correct the skew of the image, but this article does not specifically discuss that method.

When an image is obtained, the peripheral distribution of character bits in the sub-scanning direction is counted 102 for the obtained image. Figure 2 2
01 shows the counting results of the peripheral distribution in the sub-scanning direction of an image in which a frame exists. The peripheral distribution of an image where a frame exists is the frame distribution T
1. T2 and character area distribution M can be detected. In character recognition, if there is a premise that a character string near the center of the obtained image is to be recognized, the area of the character area distribution M becomes the character area to be recognized. When character recognition is performed on all regions of the obtained image, all regions in which the character region M can be detected in the image become recognition target regions. FIG. 4 401 shows the counting results of the peripheral distribution in the sub-scanning direction of an image without a frame. If there is no frame, only the character area M can be detected, so characters are recognized for that area. Next, a method 103 for determining the frame distribution T and character distribution M will be described. The character determination reference value Refl for determining whether the area is a character area is derived from the following equation.

Refl=average value of peripheral distribution frequency/3; (formula 001) average value of peripheral distribution frequency - total frequency of peripheral distribution/number of bits in main scanning direction; (formula 002
) A distribution area having a distribution frequency less than or equal to the Refl value is an area that is not a character area. The determination of a frame area and a character area can be derived from the equation shown below.

M>4XTi * Criterion formula for continuous distribution area (Formula 003) T h > Mh 1 Mh/2 * Criterion formula for distribution frequency (Formula 004) Search the image for a character area M that satisfies the above formula, Define the image area. Here, when a frame T area satisfying the above formula is detected, the frame area is erased 104 from the original image. Erasing is achieved by extracting the contours of connected components in eight directions in the determined T area and filling the inside of the extracted contour with white bits.

Next, the peripheral distribution of character bits in the main scanning direction is counted 10
Do 5. FIG. 2 203 shows the counting results of the peripheral distribution in the main scanning direction of an image in which a frame exists. The distribution count is calculated for the frame of the previously determined character line area and the area surrounded by the frame, or for the character area line if the frame area does not exist. The peripheral distribution of the image where the frame exists is the frame cloth t1. t2 and character area distribution m can be detected. FIG. 4 402 shows the counting results of the peripheral distribution in the main scanning direction of an image without a frame. If there is no frame, only character area m can be detected, and therefore characters are recognized for that area. Next, a method 106 for determining the frame cloth t and the character distribution m will be described.

The character determination reference value Ref2 for determining whether or not it is a character area is derived from the following equation.

Ref2=average value of peripheral distribution frequency/8; (formula 005) average value of peripheral distribution frequency 2 total frequency of peripheral distribution/number of pits in sub-scanning direction; (formula 006
) A distribution area having a distribution frequency less than or equal to the Ref2 value is an area that is not a character area. The determination of a frame area and a character area can be derived from the equation shown below.

m > 4. X t ; *Judgment criterion formula for continuous distribution region (formula 007) Search for true m-regions and t-regions that satisfy the following formulas in the m-regions and t-regions that satisfy the above formulas.

t h > m h + m h/2*Distribution frequency judgment criterion formula (formula 008) The image region is determined by searching the image for a character region m that satisfies the above formula. Here, when a frame area satisfying the above formula is detected, the frame area is erased 107 from the original image. Erasing is achieved by extracting the contours of connected components in eight directions in the determined t-area and filling the inside of the extracted contour with white bits.

By using the above method, it is basically possible to remove the frame component, but if the read image is input diagonally as shown in Figure 3, the frame can be removed even with the above method. less likely to be able to do so. The reason for this is that the frequencies of Th and th during marginal distribution counting become small, making it impossible to determine the frame using the above discriminant. Therefore, a two-step image removal method is adopted using the method described below.

When the character string area for character recognition can be limited using the method described above, the character recognition main routine starts. First, character recognition involves cutting out each character from the original image. Character cutting is performed by extracting outlines of connected components in eight directions of each character image. Here, before extracting the contour components of characters, an upper limit region for contour extraction and a lower limit region for contour extraction are set 108. The outline extraction upper limit area is determined based on the upper limit sub-scanning line of the character area obtained when counting the peripheral distribution in the sub-scanning direction. Its value is: contour extraction upper limit area=character area line upper limit line−character line area length/3 or image input sub-scanning line start line: (Formula 009). Further, the lower limit area for outline extraction is determined based on the =11-lower limit sub-scanning line of the character area characteristic obtained when counting the peripheral distribution in the sub-scanning direction. The value is 109 degrees in 8 directions beyond the limit area calculated by the above formula. If you attempt to perform outline extraction (111), it can be determined that the character image whose outline is being extracted is not a character, that is, a frame component in the image, or vertical line noise (a value larger than the character determination standard value). (because it is a character)
, contour extraction for that character is terminated at the moment the determination reference line is accessed, and the process moves on to contour extraction 112 for the next character. If contour extraction in eight directions is successful without accessing the judgment reference line, it can be determined that the contour-extracted character is a true character, and by character normalization 113 and matching with the pattern dictionary 114, Ending one character recognition and proceeding to recognition of the next character.The above character recognition process is repeated for all areas of the 115° recognized character area to end character string recognition.

FIG. 4 is a diagram showing an example of an input character image and an example of the distribution of marginal distribution coefficients counted for the image.

(Effects of the Invention) As described above, according to the present invention, in a base paper whose format cannot be determined, frame components surrounding characters or vertical line noise are present on the paper surface to be recognized. It also enables accurate character recognition.

that's all Applicant: Seiko Epson Corporation Agent: Patent attorney Kizobe Suzuki and 1 other person

[Brief explanation of drawings]

FIG. 1 is a diagram showing an example of a flowchart configuration according to the present invention, and FIG. 2 is a diagram showing an example of a character image input from a base paper with a frame and an example of the distribution of marginal distribution coefficients counted for that image. Figure 3 shows an example of a character image input diagonally and an example of the distribution of marginal distribution coefficients counted for that image (α (c)

Claims (1)

[Scope of Claims] (a) Character image reading means realized by an optical sensor or the like; (b) Means for counting whether the input character area is a character in the sub-scanning direction; (c) means for repeating the counting means in the main scanning direction; and (d) means for determining whether the main scanning direction component is vertical line noise or a frame surrounding a character based on the counted value in the main scanning direction. (e) means for determining an area for character recognition based on the counted marginal distribution; (f) means for extracting and normalizing characters by extracting their outlines; (g) at the time of extracting the outlines; (h) a pattern dictionary or a peripheral feature dictionary for recognizing characters; (i) Character recognition means, comprising means for comparing the normalized characters with the recognition dictionary.