JP2795860B2 - Character recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial application field) The present invention relates to a character recognition device for recognizing handwritten blocked numbers in, for example, an optical character reader.

(Prior art) Conventionally, as a method of recognizing handwritten blocked numbers,
The input character pattern is normalized, and pre-processing for creating a sampling pattern is performed by performing a predetermined process on the normalized character pattern. The sample pattern obtained by this pre-processing is prepared in advance. A method of recognizing input characters based on the similarity and the distance of the dictionary pattern being used, or scanning the area corresponding to each segment constituting the handwritten block figure in the input character pattern and the area corresponding to each segment There is known a method of recognizing an input character by determining whether or not a pixel exists.

However, the method of recognizing an input character based on the similarity and the distance is disadvantageous in that it takes an enormous amount of time to calculate the sum of products for calculating the similarity and the distance, and is inferior in recognition speed. In the case where an input character is recognized by scanning an area corresponding to a segment and determining whether or not a pixel exists in the area, the recognition rate of the input character is reduced if a line segment is cut or blurred. There was a drawback of lowering.

(Problem to be Solved by the Invention) The character recognition device of the present invention recognizes an input character based on the similarity between a sampling pattern obtained by preprocessing and a dictionary pattern prepared in advance and the magnitude of a distance. The disadvantage is that the product-sum calculation for calculating the similarity and the distance takes an enormous amount of time and the recognition speed is inferior, and the region corresponding to the segment is scanned to determine whether or not a pixel exists in the region. The recognition of input characters by doing so is to solve the disadvantage that the recognition rate of input characters is reduced if there is a cut or blurring of line segments, and it is possible to perform character recognition at high speed, It is an object of the present invention to provide a character recognition device capable of obtaining a high recognition rate even if a line segment of an input character pattern is cut or blurred.

[Structure of the Invention] (Means for Solving the Problems) A character recognition device according to the present invention is a character recognition device for recognizing a handwritten blocked digit, comprising: a normalization unit for normalizing an input character pattern; Sampling means for sampling by normalizing the output normalized pattern by blurring,
A feature extraction unit that extracts a density feature of a pixel corresponding to each segment constituting a handwritten blocked number in a sampling pattern output by the sampling unit to obtain a feature value, obtained by the feature extraction unit; Binarizing means for binarizing the feature value, and determining whether or not each segment constituting a handwritten blocked digit is present, based on the feature value binarized by the binarizing means, It consists of recognition means for recognizing patterns.

(Operation) The present invention normalizes an input character pattern given as a handwritten blocked digit, blurs the normalized pattern obtained by this normalization, obtains a sampled pattern, and obtains a sampled pattern. The pixel density of a portion corresponding to each segment constituting the handwritten blocked number is extracted as a feature value, and the presence / absence of each segment constituting the handwritten blocked number is determined based on the feature value so as to recognize the input character pattern. It was done. As a result, the amount of sum-of-products calculation is drastically reduced, enabling high-speed recognition processing, and is less susceptible to cuts or blurring of line segments.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

FIG. 2 shows an entry frame for handwritten blocked numbers to be recognized by the character recognition apparatus of the present invention. That is, a number (line segment) is entered by tracing a portion shown by a dotted line within a thick line frame, and one number is
Any of the twelve segments 11 to 22 is represented in combination.

Fig. 3 shows the numbers "1" to "0" to be entered in the above entry boxes.
When recognizing an input number, it is processed on the assumption that the input is made in accordance with the standard character shape.

FIG. 1 schematically shows the configuration of the character recognition device of the present invention. In the figure, 1 is a normalization unit. The character pattern input to the normalization unit 1 generates an image signal by photoelectrically converting image information captured by an imaging device such as a camera by a photoelectric conversion element such as a CCD (Charge Coupled Device). This image signal is given as binary data of black pixels and white pixels according to the pixel density. The normalization unit 1 having input such a character pattern normalizes the size and the character position by enlarging or reducing the vertical width or the horizontal width of the character pattern. In this normalization, when the width of the input character pattern is equal to or smaller than a certain threshold value, the width is not normalized, and only the character position is normalized so that the width of the number “1” is not excessively enlarged. Is performed. FIG. 4 (a) shows an example in which the input numeral "3" is normalized to 11 pixels × 11 pixels.

The sampling unit 2 shown in FIG. 1 receives the normalized pattern normalized by the normalizing unit 1 and performs a blur transform according to a sampling smack to obtain a sampled pattern.

In the blur conversion in the sampling unit 2, for example, as shown in FIG. 4 (b), weights (“1”, “2”, “4” in the illustrated example) are assigned to each pixel of 3 × 3 pixels. This is performed using the attached sampling mask. That is, the sampling mask is moved while being shifted vertically and horizontally by two pixels on the normalization pattern created by the normalization unit 1, and the sum of the weights of the respective pixel components of the sampling mask is calculated. To obtain the pixel density value at the position. FIG. 4 (c) shows an example of the sampling pattern thus obtained. That is,
By shifting by two pixels, a sampling pattern in which each pixel component of a matrix of 5 pixels × 5 pixels is represented by a pixel density value is obtained. Note that all the pixels in the normalized pattern surrounded by the sampling mask are “0” for the vapor pixel density value.
, The minimum value is “0”, and when all the pixels in the normalized pattern are “1”, the maximum value is “16”. In other cases, the distribution is between “0” and “16”. I do.

The feature extraction unit 3 shown in FIG. 1 receives the sampling pattern from the sampling unit 2 and calculates the pixel density corresponding to each of the segments 11 to 22 of the handwritten block numbers based on the sampling pattern. It is. That is, assuming that each pixel component of a matrix of 5 pixels × 5 pixels is a to y as shown in FIG. 5 (a), the weight of the center position of each segment is “1”, The pixel density for each segment is calculated with the weight set to “1/2”, and this is set as the characteristic value of the segment. For example, the feature value of segment 11 is
It is calculated by the formula of “a / 2 + b + c / 2”. The characteristic values of the other segments are calculated by the same formula as shown in FIG. 5 (b). The region where no segment exists, that is, the positions of g, i, q, and s in FIG. 5 (a) is a blank portion, and is usually a portion where the pixel density becomes zero. Therefore, it is a portion where the probability that a pixel appears is small. Therefore,
This part is considered to have a larger feature than the other parts, and the feature extraction is performed with the weight being “2”. With the 16 feature values obtained as described above,
The following number recognition will be performed.

The binarizing unit 4 in FIG. 1 binarizes the feature value extracted by the feature extracting unit 3 by slicing it at a predetermined slice level to obtain a binarized feature value. In this embodiment, the slice level is set to “12”.

The recognizing unit 5 includes a binarized characteristic value obtained by the binarizing unit 4 and a reference prepared in advance as binarized characteristic values of the standard characters “1” to “0”. The input character pattern is recognized by comparing with the characteristic value. FIG. 6 shows the correspondence between the types of standard feature values prepared in advance and the numbers output as recognition results.

Next, the operation in the above configuration will be described.

First, an image signal is obtained by taking an image of a block number written in the entry frame shown in FIG. 2 by handwriting with an imaging device (not shown) and performing photoelectric conversion. Then, the image signal is binarized by associating black pixels with “1” and white pixels with “0”, and input to the normalization unit 1 as a character pattern to be recognized.

The normalizing unit 1 receives the character pattern and normalizes the size and position. FIG. 4A shows an example of a character pattern when the number “3” is normalized to 11 × 11 pixels. In the figure, the portion indicated by "1" is a black pixel portion, and the portion indicated by "0" is a white pixel portion.

The normalized pattern created in this way is sent to the sampling section 2. The sampling unit 2 performs blur conversion on the sampling pattern to obtain a sampling pattern. That is, the sampling mask shown in FIG. 4B is projected onto the normalization pattern shown in FIG. 4A, and the pixel density of the area is calculated according to the weight defined by the sampling mask. For example, the pixel density when a sampling mask is projected on the upper left corner 30 of the normalized pattern is “1 × 1 + 1 × 2 + 1 × 1 + 1 × 2 + 1 × 4”.
+ 1 × 2 + 0 × 1 + 0 × 2 + 0 × 1 = 12 ”. Next, the sampling mask is shifted rightward by two pixels, and the pixel density “12” is obtained in the same manner. In the same manner, the pixel density is sequentially obtained in the right direction, and when the pixel density "15" at the upper right corner is obtained, the pixel density is shifted downward by two pixels. Then, the pixel density of the second column is obtained by the same calculation as above. When the pixel density “15” at the right corner of the bottom row is obtained in this way, the blur conversion processing ends.

The sampling pattern obtained by the above process is obtained as a matrix as shown in FIG. 4 (c), and the pixel pattern at each pixel component position when the input character pattern is represented by a matrix of 5 pixels × 5 pixels. It reflects the pixel density. In other words, when the matrix of 5 pixels × 5 pixels is viewed in association with each segment forming the handwritten blocked number, the segments 11, 12, and
It can be seen that the pixel density values in the vicinity corresponding to 15, 16, 17, 20, 21, and 22 are large.

The sampling pattern created in this way is sent to the feature extraction unit 3. The feature extraction unit 3 extracts features of the sampling pattern. That is, the feature value is calculated for each segment using the above-described formula. FIG. 5B shows the characteristic values of each of the segments 11 to 22 calculated according to the above formula. At this time, the feature value corresponding to the area where the segment of the handwritten blocked number does not exist, that is, the area of g, i, q, s in FIG. 5A is also calculated according to the formula shown in FIG. 5B. .

Next, the feature value is sent to the binarization unit 4. The binarizing unit 4 binarizes the feature value by slicing it at a slice level “12” to obtain a binarized feature value.

The binarized feature value obtained by the binarization unit 4 is sent to the recognition unit 5. The recognizing unit 5 performs a process of sequentially comparing the reference characteristic values of the numbers “1” to “0” as the standard character shapes prepared in advance. Then, when there is a matching reference feature value, a numeral (answer) corresponding to the reference feature value is output as a recognition result. On the other hand, if there is no matching reference feature value, it is rejected as unrecognizable.

As described above, the input character pattern given as a handwritten blocked number is normalized, and a blurring transformation is performed on the input character pattern to obtain a sampling pattern. The feature is extracted from the pixel density of the corresponding part, and the input character pattern is recognized by judging the organic character of each segment constituting the handwritten blocked number based on the feature value. The amount of sum-of-products calculation is dramatically reduced as compared to
As in the case where a region corresponding to a segment is scanned and character recognition is performed based on the presence or absence of the segment, it is hardly affected by cuts or blurring of line segments, so that the recognition rate can be improved.

In the above-described embodiment, when the feature value obtained by the feature extraction unit 3 is binarized, the thrust level is set to “12”. However, the present invention is not limited to this, and another value may be used.

[Effects of the Invention] As described above, according to the present invention, character recognition can be performed at high speed, and a high recognition rate can be obtained even if a line segment of an input character pattern is cut or blurred. A character recognition device can be provided.

[Brief description of the drawings]

FIG. 1 shows an embodiment of the present invention. FIG. 1 is a block diagram schematically showing the entire structure, FIG. 2 is a diagram showing an entry frame for handwritten blocked numbers, and FIG. FIG. 4 is a diagram showing a standard character shape of numbers, FIG. 4 is a diagram for explaining the sampling operation, FIG. 4A shows an example of an input normalization pattern, and FIG. FIG. 5C is a diagram showing an example of a sampling mask, FIG. 5C is a diagram showing an example of an output sampling pattern, FIG. 5 is a diagram for explaining an operation of feature extraction, and FIG. FIG. 6B is a diagram showing the position of each pixel component of the conversion pattern, FIG. 6B is a diagram showing a feature value calculation formula for each segment, a calculated feature value and a binarized feature value, and FIG. FIG. 6 is a diagram showing a relationship between reference feature values and recognition results. 1 ... Normalizing section (Normalizing means), 2 ... Sampling section (Sampling means), 3 ... Feature extracting section (Feature extracting means), 4 ...
Binarization unit (binarization means), 5 ... recognition unit (recognition means),
11-22 …… Segments.

Claims (1)

(57) [Claims] 1. A character recognition apparatus for recognizing a handwritten blocked digit, comprising: a normalizing means for normalizing an input character pattern; and a sampling means for performing sampling by subjecting the normalized pattern output by the normalizing means to blur conversion. Means for extracting a density feature of a pixel corresponding to each segment constituting a handwritten blocked number in a sampling pattern output by the sampling means to obtain a feature value; and a feature extracting means. Binarizing means for binarizing the feature value obtained by the above, and judging the presence or absence of each segment constituting a handwritten blocked digit based on the feature value binarized by the binarizing means. A character recognition device for recognizing the input character pattern.