JPS5942906B2 - Line width extraction device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a line width extraction device, and in particular, in a character recognition method, the line width of a character to be recognized, such as a handwritten character to be recognized, remains almost unchanged even when the size of the character changes. This invention relates to a line width extraction device that extracts the average character line width of the recognition target character in order to use it as one of the feature extraction criteria of the recognition target character.

Generally, in a character recognition device, recognition is performed by extracting the structural features of a character to be recognized, but when extracting the features, the size of the character to be recognized is normalized to perform feature extraction.

That is, for example, when performing feature extraction to determine whether vertical strokes divided into two by a horizontal stroke form one continuous vertical stroke in the katakana "su", the lower Y coordinate of the upper vertical stroke and the lower When the difference from the upper end Y coordinate of the vertical stroke is less than a predetermined threshold value, it is determined that the continuous vertical stroke is formed. However, when the recognition target character is a handwritten character, the line width of the character remains the same regardless of the size of the character if the same type of writing instrument is used. For this reason, for example, if a character written in small size is enlarged for normalization, the character line width of the normalized character becomes undesirably large as will be described later with reference to FIG.
Therefore, in the case of the katakana "su", the vertical stroke may not be determined as one continuous vertical stroke. For this reason, even if an attempt is made to make the above judgment without normalizing the size of the characters, it will not provide an essential solution, as will be described later with reference to FIG.

For this reason or for other reasons, it is necessary to perform processing to check the character line width of the character to be recognized. An object of the present invention is to solve the above-mentioned problems, and to detect the line width of the line image by scanning a line image existing in a predetermined area and detecting white mesh points and black mesh points. A line width extraction device for extracting a line width, the black mesh point counting means for counting the number of consecutive black mesh points that appear continuously during the scanning, and the output of the black mesh point counting means as the continuous black mesh point. continuous black mesh point number counting means for classifying the number of consecutive meshes according to the number of consecutive meshes and counting the value for each number of meshes; The maximum value of the number of continuous black mesh points is determined based on the output of each continuous black mesh point number counting means provided for each line, and the maximum value is output as the line width of the line image existing within the predetermined area. This object is achieved by providing a line width extraction device characterized by comprising a value determining means.

This will be explained below with reference to the drawings. 1 and 2 are explanatory diagrams for explaining the threshold value of character line width in feature extraction in a general character recognition method, FIG. 3 is an explanatory diagram for explaining the present invention, and FIG. The figure shows one embodiment of the line width extracting device used in the present invention, and FIG. 5 shows an example of a histogram obtained by scanning the character to be recognized in the horizontal and vertical directions.

As mentioned at the beginning of this specification, when examining the continuity of vertical strokes for katakana [o], the difference between the lower Y coordinate of the vertical stroke I and the upper Y coordinate of the vertical stroke I shown in Figure 1 is a predetermined threshold value. A check is made to see if it is smaller than K.

However, when handwritten characters written in a small size are normalized as described at the beginning of this specification, the character line width of the normalized characters becomes undesirably large as shown in FIG. As a result, the vertical stroke and the continuity of the mouth are lost. For this reason, it is conceivable to check the continuity of vertical strokes and openings before normalizing the size of characters as shown in FIGS. 1 and 2. However, the threshold value K in such cases is generally determined based on the size of the characters, and in the case of FIG. 2, the threshold value K1 is used,
In the case of FIG. 2, the threshold value K2 is used.
Therefore, contrary to the case shown in FIG. 1, the continuity of vertical strokes and mouths is lost in large handwritten characters. FIG. 4 shows an embodiment of the present invention in which an approximate character line width of the recognition target character is extracted based on the result of scanning the recognition target character in consideration of the above points.

In the figure, I is a character line width classification processing unit during horizontal scanning, which is a part of the line width extraction device of the present invention, or a character line width classification processing unit during vertical scanning, and a is a white mesh point during horizontal scanning. b is a signal indicating that the continuous black mesh point starting from a has shifted to the white mesh point for the first time.

and d are signals corresponding to the above a and b during vertical scanning, A is a flip-flop that switches continuous black mesh point information, and B is a flip-flop that counts the number of black meshes of continuous black mesh points from the output of flip-flop A. Counter C is a decoder, D1 to DO, which classifies the output of counter B according to the number of black meshes of the continuous black mesh points.
is an AND circuit, E1 to En are continuous black mesh point number counters that count the output of decoder C according to the number of black meshes 1 to n of continuous black mesh points, and H1 to HO are the outputs of counters E1 to En, which are used in horizontal scanning. Character line width number information indicating that the number of black meshes of continuous black mesh points is 1 to n, V1 to VO correspond to the above H1 to Hn when the recognized character is scanned in the vertical direction, and F1 to Fn are the above A counter that adds the outputs H1 to Hn and V1 to Vn obtained as a result of scanning the recognition target characters in the horizontal and vertical directions, respectively, that is, a counter that counts the above-mentioned H1+V1 to Hn+v1, and G is the value of one of the recognition target characters. When all scanning is completed, outputs T1 to T of counters F1 to Fn
Each represents a maximum value determining circuit that determines the maximum value of n. 4. In FIG. 4, the character to be recognized is scanned in the horizontal direction, and a signal a indicates that the character has moved from the white mesh point to the first black mesh point that appears (for example, a1 or A2 in the H scanning line in the example shown in FIG. 3). The flip-flop A is set by the signal b (for example, b1 or B2 in the H scanning line in the example shown in FIG. 3) indicating that the black mesh point that appears continuously thereafter has shifted to the white mesh point that appears for the first time. Flip Flop A
The output of flip-flop A, that is, the number of meshes of black mesh points that appear continuously during scanning (
For example, the number of black meshes between a1 and b1 and between A2 and B2 in the H scanning line shown in FIG. 3 is counted by counter B.

Note that the counter B receives the signal a before starting counting.
It is set by The output of counter B, which counts the number of meshes of consecutive black mesh points, is sent to decoder C.
Accordingly, the continuous black mesh points are classified into mesh numbers 1 to n. The outputs classified by the decoder C are sent to counters E1 to En via AND circuits D1 to DO, respectively, and the continuous black mesh points are classified into mesh numbers 1 to n, and are sent to the counters E1 to En.
Counted from 1 to En. In this way, the information regarding the continuous black mesh points obtained by scanning the character to be recognized in the horizontal direction is classified according to the number of meshes of the black mesh points and counted by counters E1 to E1. The recognized character is then vertically scanned and the same means as described above are applied to obtain the outputs V1 to Vn of the circuit shown in FIG. And the above output H1 to H
n and outputs 1 to V2 are counters E1 to Fn, respectively.
sent to. 4th in counters F1 to FO, respectively.
Outputs H1 to Hn sent from Figures I and 4
The outputs T1 to Tn obtained by adding V1 to Vn, that is, H1+V1 in the counter F1 and H2+V2 in the counter F2, are sent to the maximum value determination circuit G. In the maximum value determination circuit G, the outputs T1 to Tn of the counters F1 to Fn, that is, the continuous black mesh points that appear during the entire horizontal and vertical scanning of one character to be recognized, are calculated as the continuous width of the black mesh points. The maximum value of the histogram is determined based on the histogram as shown in the example of FIG. 5, based on the number of consecutive black mesh points classified and totaled from 1 to n. The approximate character width of the character to be recognized is determined using the histogram maximum value given by the maximum value determining circuit G. That is, in general, for line images such as characters (particularly kanji) and figures, a predetermined area (for example, in the case of characters, an area of only one character after segmentation)
Most of the strokes in , extend in the horizontal and vertical directions, and by scanning a predetermined area in the horizontal and vertical directions, a histogram of the lengths of consecutive black mesh points in the scanning direction is obtained. The length can be an approximate character line width. As explained above, by using the line width extraction device of the present invention, the instability of feature extraction caused by variations in the size of handwritten recognition target characters can be solved, and By focusing on the fact that the character line width of the recognition target character is almost constant, it is possible to extract an approximate character line width of the recognition target character. Therefore, by using this approximate line width as a threshold for determining the continuity of vertical strokes, for example, it is possible to avoid incorrect feature extraction when extracting more structural features of the recognition target character. (Determining that the vertical strokes are not continuous even though they are continuous) is no longer performed, making it possible to correctly recognize characters.

[Brief explanation of the drawing]

1 and 2 are explanatory diagrams for explaining the threshold value of character line width for feature extraction in a general character recognition method; FIG. 3 is an explanatory diagram for explaining the present invention; FIG. 4 shows an embodiment of the line width extraction device used in the present invention, and FIG. 5 shows an example of a histogram obtained by scanning the character to be recognized in the horizontal and vertical directions. In the figure, I is a character line width classification processing unit during horizontal scanning, character line width classification processing unit during vertical scanning, A is a flip-flop,
B is a counter, C is a decoder, D1 to DO are AND circuits, E1 to En and F1 to F are respective counters, a is a signal indicating that the mesh point has shifted from white to black during horizontal scanning, b is from a Signals indicating that the starting continuous black mesh point has shifted to the white mesh point, c and d correspond to the above a and b during vertical scanning, H1 to Hn are the number of meshes of character line width during horizontal scanning, 1 The number of character line widths classified and totaled from n to n, V1 to Vn are the above H1 to Hn during vertical scanning.
, T1 to Tn represent the addition numbers of H1+V and Hn+Vn, and G represents the maximum value determining circuit, respectively.

Claims (1)

[Claims] 1 A line width extraction device that extracts the line width of a line image by scanning a line image existing within a predetermined area and detecting white mesh points and black mesh points, the device comprising: A black mesh point counting means for counting the number of consecutive black mesh points that appear continuously and the output of the black mesh point counting means are classified according to the number of consecutive black mesh points, and a value is calculated for each number of meshes. a continuous black mesh point number counting means for counting the number of continuous black mesh points is provided for each scanning line that scans the predetermined area in at least two substantially orthogonal directions, and an output of each continuous black mesh point number counting means provided for each scanning line is provided. and a maximum value determining means for determining the maximum value of the number of continuous black mesh points based on the number of continuous black mesh points and outputting the maximum value as the line width of the line image existing within the predetermined area. Device.