JPS60181883A - Stroke extraction method in character discrimination - Google Patents

Abstract

PURPOSE:To extract a stable, high speed stroke by deleting unnecessary segment based on connecting point coordinates extracted from four directional subpattern or intersection coordinates, prolonging only efficient segments, and synthesizing segments connected at an end point. CONSTITUTION:A vertical segment detecting parts 5a calculates a coordinate location of a black point continuous area from the black point coordinate at the time of a change from a white point (character background part) to the black point (character line part) and from the white point coordinates at the time of the change from the black to the white points, detects the number of such locations at the change point, and then detects a center point coordinate system in the black point continuous area as a vertical primitive segment. Equally, it detects a horizontal, left oblique, and right oblique primitive segments. An efficient segment detecting part 9 deletes unnecessary segments of plural segments detected by the stroke having four directional intermediate angles, and when the storke concentrates, the part 9 deletes unnecessary segments detected at connecting point/intersections. A stroke extracting part 11, when the efficient segment derived from an efficient segment table memory 10 can be prolonged to the other efficient segments, then said segments are synthesized to obtain one efficient segment.

Description

DETAILED DESCRIPTION OF THE INVENTION (Technical Field) The present invention relates to a character recognition device, particularly a character recognition device for handwritten characters. Regarding the stroke extraction method.

(Background Art) Conventional methods for extracting strokes include (1) calculating the curvature of a series of contour points detected by tracing the contour of a character figure, and dividing the contour series by using points with large values of curvature as dividing points; There are two methods: extracting strokes by dividing and combining the divided series, or (2) performing thinning processing on the character figure pattern to create a skeleton, tracing the connectivity and skeleton pattern of the skeleton pattern, and then extracting strokes by combining the divided series. There is a method of extracting strokes by detecting points of change in angle, etc.

However, method (1) loses track of the stroke direction when strokes are concentrated and extracts strokes in the wrong direction.Furthermore, as the character/figure pattern becomes larger and more complex, the amount of processing increases, which reduces the processing speed. In method (2), since it is necessary to thin the character/figure pattern, distortion of the pattern due to thinning, whiskers, etc. occur, and if strokes are concentrated, it is necessary to misunderstand the connectivity of the pattern. However, there was a drawback that the stroke could not be extracted stably.

(Problems to be solved by the invention) The present invention improves the above-mentioned drawbacks of the conventional techniques and provides a strokeless extraction method that enables stable and high-speed stroke extraction. ρ Based on the connection point coordinates or intersection coordinates of the extracted segments, unnecessary segments are deleted, only valid segments are extended, segments connected at the end points are combined, and strokes are extracted. .

(Structure and operation of the invention) FIG. 1 is a block diagram showing one embodiment of the present invention.

1 is a binary pattern input signal, 2 is a pattern register, 3
4a is a line width calculation section, 4a is a vertical sub-pattern extraction section, 41]
4C is a horizontal subpattern extraction section, 4C is a left diagonal subpattern extraction section, 4d is a right diagonal subpattern extraction section, 5a is a vertical segment extraction section, 5b is a horizontal segment extraction section, 5C
5d is a left diagonal segment extraction unit; 5d is a right diagonal segment extraction unit; 6 is a primitive segment table memory that stores the primitive segments detected by each segment extraction unit; 7 is the coordinates of connection points between the detected primitive segments; a connection point intersection detection unit that detects intersection coordinates; 8 is a connection point coordinate;
Connection point/intersection table memory that stores intersection coordinates,
Reference numeral 9 denotes a valid segment detection unit that detects valid segments from the detected connection points and intersection table memory; 1O denotes a valid segment table memory that stores the detected valid segments; and 11, synthesizes the detected valid segments and generates a stroke. A stroke extractor 12 indicates a stroke table memory that stores extracted strokes.

The binary pattern input signal 1 is stored in the pattern register 2. At the same time, the line width of the input pattern is calculated in the line width calculation section 3. The vertical sub-pattern extraction section 4aH performs vertical scanning on the pattern register 2, extracts the continuous black bits whose length is sufficiently longer than the line width calculated in the line width calculation section 3, and extracts the continuous black bits from the vertical sub-pattern extraction section 4aH. (
Extract the turn (VSP). Similarly, the horizontal sub pattern (H4
F), the left diagonal sub-pattern extraction unit 4C extracts the left diagonal sub-pattern (LSF) by scanning diagonally 45 degrees to the left, and the right diagonal sub-pattern extraction unit 4d extracts the right diagonal sub-pattern (H8F) by scanning diagonally 45 degrees to the right. do.

Figure 2 is a diagram showing an example of extracted sub-patterns of the characters r L J. Figure 2 (a) is the original pattern, Figure 2 (b) ( is the horizontal sub-pattern, Figure 2 (C) is the vertical sub-pattern, Figure 2 (I) is the left diagonal sub-pattern, Figure 2 (I) is the left diagonal sub-pattern, Although (e) is a right diagonal subpattern, the right diagonal subpattern is not extracted in this example.

Next, the vertical segment detection unit 5a detects the vertical sub-pattern (V
SP) horizontally from left to right to find the white point (character background)
Calculate the coordinate position of the continuous area of the black point from both the black point coordinates when it changes from a black point to a black point (character line part) and the white point coordinates when it changes from a black point to a white point, and then calculate the number of changing points. To detect. This operation is performed from top to bottom, and the center point coordinate series of the continuous region of black points is detected as a vertical primitive segment from the coordinate positions of the continuous region of black points in the previous row and the current row and the number of changing points. Similarly, the horizontal segment detection unit 5b scans the horizontal sub pattern (H8PI from top to bottom to detect horizontal primitive segments), and the left diagonal segment detection unit 5c scans the left diagonal sub pattern (LSP) from left to right. The left diagonal primitive segment is detected, and the right diagonal segment detection unit 5d scans the right diagonal sub pattern (H8F) from left to right to detect the right diagonal primitive segment. Patent application already filed 1982-3
1228 (Japanese Unexamined Patent Publication No. 57-146379), the method will be explained by taking as an example the case of detecting vertical primitive segments by performing core line processing on vertical subpattern (VSP) components. do.

Now, assume that the vertical subpattern (VSP) has a pattern of J-shaped segment components A and B shown in FIG. 3. When starting from the upper left point and scanning from left to right to detect change points, no change points are detected until the line before the Y1 line, and it is only after scanning the Y1 line that the two change points of stroke component A are detected. That is, a change point XBYl from a white point to a black point and a change point XwY1 from a black point to a white point are obtained. At this time (XBYI + XW
YI)/2 is calculated and this coordinate is registered in the primitive segment table memory 6 as the starting point of the primitive segment 1.

Furthermore, the two change points obtained by scanning the next row are XBY1+1. Expression (1) shown below as Xw, □10□
, when equation (2) is satisfied, it is defined that the coordinate position of the continuous area of the currently detected black point overlaps with that of the black point detected in the previous line, and is considered to be a continuation of the segment of the previous line.

XBYl≦XWY1+1 ・・・・・・ (1) XWY
1≧XDY1+1 ・・・・・・ (2) (However, XB
Y engineering〈XwY□, XBY1+□〈XwY1+□.

) In other cases, the center coordinate position of the continuous area of black points detected in the current row is registered in the primitive segment table memory 6 as belonging to a primitive segment different from the previous row.

In this way, scanning is performed up to the 72nd previous line, and the coordinates of the center point of primitive segment 1 are registered. When the 72nd row is scanned, the number of change points becomes 4Sl, indicating that a new segment component B has appeared. By repeating the operations described above,
There are two types: one is registered as a continuation of primitive segment 1, and the other is registered in the primitive segment table memory 6 as the starting point of a new primitive segment 2 whose center point coordinate position is determined separately from primitive segment 1. In this way, it scans up to the line before the \3 line and selects segment component A.

The coordinates of the center point of B are respectively registered in the primitive segment table memory 6. When the Y3 column is scanned, there are two changing points, and it can be seen from the coordinate positions of the continuous area of the black dots that it is a continuation of the primitive segment 2, indicating that the detection of the primitive segment 1 has been completed.

There are two changing points from this line to the previous line of Y4 line, and it is registered as a continuation of original segment 2. When the Y4 row is scanned, there are no changing points, which means that the detection of the original segment 2 has been completed, and that all segment extraction of the vertical sub pattern (VSP) has been completed. A similar function is provided by the horizontal segment detection section 5b.

The detection is performed by the left diagonal segment detection section 5C1 and the right diagonal segment detection section 5d, and the detected primitive segments are stored in the primitive segment table memory 6, respectively.

FIG. 4 shows the original segment extracted by converting the subpattern of the letter "L" shown in FIG. 1 into core lines as described above. Figure 4(a) shows the horizontal primitive segment, Figure 4(1)) shows the vertical primitive segment, Figure 4(C
) indicates a left diagonal primitive segment.

The connection point/intersection coordinate detection unit 7 detects connection/intersection relationships for all the primitive segments stored in the primitive segment table 6, and stores them in the connection point/intersection table memory 8.
The operation flowchart 1 is to store the results in
・is shown in Figure 5. First, one primitive segment (referred to as primitive segment 1) is taken out from the primitive segment table memory 6, and the distance between each point of another primitive segment (referred to as primitive segment 2) and each point of the retrieved primitive segment 1 is , and detect the coordinates of each point in primitive segment 1 and primitive segment 2 (hereinafter referred to as connection/intersection candidate points) whose distance is less than or equal to C, (a, a constant of 1 in this example), and connect/intersect candidate points. When an intersection candidate point exists, primitive segment 1 and primitive segment 2 are considered candidates for connection/intersection. Whether it is a connection or an intersection is determined by the following process.

Normally, if there are connection/intersection candidate points, multiple points are consecutive, so primitive segment 1. For each primitive segment 2, detect the interval between consecutive connection/intersection candidate points within the self-segment, calculate the distance from each of the starting point and end point to both end points of the self-segment, and calculate this distance as C2W (C2 is a constant). is 1 in this example, and W is the line width) In the following case, primitive segment 1 and primitive segment 2 are connected (@ connection/intersection candidate point is determined to be a connecting point). At the same time, primitive segment 1. For each primitive segment 2, the coordinates of the midpoint of consecutive connection points of the segment and information on whether it is connected to the start point or end point of the segment are stored in the connection point/intersection table memory 8. Other than this, it is determined that primitive segment 1 and primitive segment 2 are intersections (the connection/intersection candidate point marked by @ is determined to be an intersection), and at the same time, primitive segment lI For each primitive segment 2, continuous intersections of the segments The coordinates of the midpoint of are stored in the connection point/intersection table memory 8.

The valid segment detection unit 9 deletes unnecessary segments by the following two types of processing.

Unnecessary segment deletion processing 1: For multiple segments detected in strokes with intermediate angles in four directions,
Delete unnecessary segments.

Unnecessary segment deletion process 2: Unnecessary segments detected when strokes are concentrated or at connection points/intersections are deleted.

The operation flowchart of unnecessary segment deletion processing 1 is shown in the sixth section.
FIG. 7 shows an operation flowchart of unnecessary segment deletion processing 2, and each processing will be explained in order below.

The operation of unnecessary segment deletion processing 1 will be explained using FIG. When the flag indicating an unnecessary segment is set in the connection point/intersection table 8, one valid segment (valid segment 1, H segment in Fig. 8) is selected from the small segment C (hereinafter referred to as valid segment). is taken out, an effective segment intersecting with effective segment 1 (valid segment 2, R segment in FIG. 8) is detected, and the starting point of effective segment 1 (H segment in FIG. 8) is detected.
Drop perpendicular lines from the starting point) and the ending point (I (starting point) in Figure 8 to the effective segment 2, respectively, and when the perpendicular lines intersect with the effective segment 2, calculate the respective distances, and the distances are both 04W (0, is a constant In this example, it is 2,
(W is the line width) In the following cases, a flag indicating that the valid segment 1 is an unnecessary segment is set in the connection point/intersection table memory 8.

The operation of unnecessary segment deletion processing 2 will be explained using FIG. 9. One valid segment (valid segment 1, segment ■ in Figure 9) is taken out from the connection point/intersection point table memory 8, and all valid segments that intersect or connect to that valid segment are detected. Only two valid segments are retrieved that are both connected to or intersect with at least one other detected valid segment. In Figure 9, the H1 segment and H2 segment are connected to the Hv upper segment.

The intersecting L segments are retrieved.

Next, a combination in which a triangle is formed by the effective segment 1 and the two extracted effective segments is detected, and (ΔABC, ΔA B'O' are detected in FIG. 9), the effective segment 1 is detected. The coordinates of the midpoint of the point C (point A, point B, or point A, point B' in FIG. 9) where the other two effective segments connect or intersect, respectively (midpoint or D' in FIG. 9) point).

Furthermore, the coordinates of the point where the two effective segments connect or intersect (point C or point 07 in Figure 9) are calculated, and the distance from the calculated midpoint is C in Figure 9. Calculate the distance between points - D or between points 07 and D', and if the distance is 0.
If it is 3W or less, the points between the two valid segments in valid segment 1 that are connected to or intersect with each other are selected as deletion candidates. The above processing is performed on the detected 3
After executing the process for all the squares, the area between each end point of effective segment 1 and the distance C2W is selected as deletion candidates.

If all valid segments 1 are deletion candidates, valid segments 1 are deleted as unnecessary segments. Valid segments other than this are valid segment table memory 10
Register to.

The stroke extraction unit 11 sequentially extracts valid segments from the valid segment table memory 10, and when the valid segments can be extended to other valid segments, extends the process of combining them with other valid segments to form one valid segment. This operation is performed until there are no more possible valid segments, and the remaining valid segments are stored as strokes in the stroke table memory 12. The operation flow is shown in FIG. The operation will be explained in detail below.

Valid segments for which the used flag is not set (valid segment 1) from the valid segment table memory 10
), and detects another valid segment (denoted as valid segment 2) whose end points are connected to valid segment 1 and whose used flag is not set.
Effective segment 1 and effective segment 2 are combined to form effective segment 1. At the same time, a flag indicating that valid segment 2 has been used is set in valid segment table memory 10. The above process is executed until there are no valid segments whose end points are connected to valid segment 1 and whose used flags are not set, and valid segment 1 is registered in the stroke table memory 12 as a stroke.

FIG. 10(a) is an example in which an effective segment of hiragana "'" is obtained, and FIG. 10(b) is an example in which two strokes are extracted from the obtained target segment.

(Effects of the Invention) As explained above, in the present invention, unnecessary segments detected at connection points and intersections, unnecessary segments detected at strokes having intermediate angles in four directions, Furthermore, unnecessary segments that occur when strokes are concentrated are deleted, remaining valid segments are extended, and strokes are extracted, so the intersecting relationships between connections between strokes become clear.

Furthermore, when extending segments, there is no need to perform complex calculations such as angles between segments, and instead focus only on the end points of valid segments, which has the advantage of fast processing speed, resulting in stable and fast character recognition. It can be used for equipment.

[Brief explanation of the drawing]

FIG. 1 is a professional diagram showing one embodiment of the present invention, and FIG.
Figures (a) to (e) are diagrams showing examples of sub-pattern extraction;
Figure 3 is a diagram explaining the stroke extraction method, Figures 4 (a) to (C) are diagrams showing segments extracted from the sub-patterns in Figure 2, and Figure 5 is connection point/intersection detection. 6 and 7 are flowcharts of the operation in the effective segment detection section 9, and FIG.
9 and 9 are diagrams for explaining the effective segment detection method, FIGS. 10(a) and (b) are diagrams for explaining the stroke extraction method, and FIG. 11 is an operation flowchart of the stroke extraction unit 11. . 1... Binary pattern input signal, 2... Pattern register, 3... Line width calculation section, 4a... Horizontal sub-pattern extraction section, 4b... Vertical sub-pattern extraction section, 4C
...Left diagonal sub-pattern extraction section, 4d... Right diagonal sub-pattern extraction section, 5a... Horizontal segment extraction section,
5b... Vertical segment extraction section, 5C... Left diagonal segment extraction section, 5d... Right diagonal segment extraction section,
6... Primitive segment table memory, 7...
Connection point/intersection detection unit, 8... Connection point/intersection table memory, 9... Valid segment detection unit, 10.
... Effective segment table memory, 11... Stroke extraction unit, 12... Stroke table memory. Patent Applicant Oki Electric Industry Co., Ltd. Patent Application Agent Patent Attorney Megumi Yamamoto - Act 2 Figure (θ) (b) (d) (e) Act 3 Figure Loaf] 1 Written amendment (voluntary) August 1980 / (Mr. Manabu Shiga, Commissioner of the Japanese Patent Office1, Indication of the case 1982 Patent Application No. 361102 Name of the invention Stroke extraction method in character recognition 3, Relationship with the person making the amendment case Patent applicant/name ( 029) Oki Electric Industry Co., Ltd. 5, Detailed description of the invention column and Brief description of the drawings column of the specification subject to amendment, and Drawing (2) [ on page 13, line 5 and line 17 of the same page] Do:ta
'J between 1']. (31 Amend “valid” in the 5th line of page 14 of the same to 1 “further valid.”) (4) “Tooshi” in the 4th line of the 15th, line 11 of the same page.
Correct it with "doshi". (5) Correct "horizontal" in line 7 of page 17 and line 10 of page 10 to "vertical". (6) Correct "vertical" to "horizontal" in line 8 of page 17 and line 11 of page 11. (7) Please refer to Figures 6, 7 and 11 of the drawings.)
2 Correct as shown in the attached sheet. that's all

Claims (1)

[Claims] A process of converting the brightness of cells constituting a character figure pattern into electrical signals and storing them in a memory, and a sub-pattern that scans the stored pattern in a plurality of predetermined directions and generates stroke components corresponding to the valley scanning direction. , a process of extracting primitive segments obtained by skeletonizing the stroke components of each sub-pattern, and detecting and removing unnecessary segments using the skeleton coordinate series of the skeleton segments, and creating effective segments other than unnecessary segments. a step of obtaining segments; a step of combining two valid segments when a condition that the end points of the two segments are connected holds; a step of repeating the step of combining the two segments until the condition no longer holds; A method for extracting strokes in character recognition, the method comprising the following steps: 1. A method for extracting strokes in character recognition, comprising the following steps: 1. A method for extracting strokes in character recognition.