JPS5939072B2 - End point detection method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention provides an end point detection method, in particular, an accurate detection method based on the original figure information obtained by scanning the figure to be recognized, without applying thinning processing to the figure to be recognized. The present invention relates to an end point detection method that detects end points at the same time.

In graphic recognition devices, particularly in recognizing handwritten characters, the end points of character strokes are extracted as one important feature.

In order to extract such end points, a so-called thinning process is conventionally performed to extract the skeleton of the recognition target figure, and the thinned recognition target figure (thinned figure) is placed in an observation window of, for example, an nXm mesh. Then, the end points were extracted. However, if the stroke width, that is, the character line width, of the figure to be recognized (character) is thick, the thinned figure obtained by the above thinning process does not necessarily faithfully represent the skeleton of the original figure. Otherwise, so-called whiskers may occur, and undesired end points may occur at unexpected locations.

In consideration of the above points, the present invention correctly detects end points by using the original figure information as is without performing line thinning processing, without employing error-prone means such as creating whiskers. The object of the present invention is to detect end points, which have been conventionally recognized as one of the important features in figure recognition.

Therefore, the end point detection method of the present invention extracts features based on the scanning information obtained by scanning the recognition target figure, and in a pattern recognition device that performs recognition processing of the recognition target figure, the recognition target figure is horizontally When scanning, each of the two figure strokes that intersect the horizontal scanning line (・
a singular line segment extraction unit that extracts a singular line segment during horizontal scanning by determining a point at an intermediate position of the width of each graphic stroke; a first area surrounded by two graphic strokes that intersect with the horizontal scanning line; An end point candidate point extraction setting unit that detects both end points of the singular line segment during horizontal scanning and extracts them as end point candidate points, and when the recognition target figure is scanned in the vertical direction, two figure strokes intersect with the vertical scanning line. A singular line segment extraction unit that determines a point at the middle position of the width of each figure stroke and extracts a singular line segment during vertical scanning; an end point candidate point extraction setting section that detects the second region of the vertical scanning singular line segment and extracts both end points of the singular line segment as end point candidate points; The present invention is characterized in that the endpoint candidate points corresponding to the overlap region where the region overlaps with the second region are not regarded as endpoints, but the remaining endpoint candidate points are detected as endpoints.With reference to the drawings below, Fig. 1 A, B, and C are explanatory views for explaining aspects of the end point detection method according to the present invention, and Fig. 2 A, B, A'', B', and C are explanatory diagrams illustrating an embodiment of the end point detection method according to the present invention. FIG. 3 is an embodiment of the configuration of the main part of the present invention for realizing the end point detection method of the embodiment shown in FIG. 2, and FIG. Explanatory diagrams are shown to explain the process of checking the inclination angle of the extending graphic stroke.In FIGS. 1A, B, and C, 1 indicates the figure (character) to be recognized, 2 indicates the figure (character) frame, 3
are horizontal scanning lines, 4 and 5 are first regions surrounded by two figure strokes intersecting the horizontal scanning direction, 6 and 7 are openings of first regions 4 and 5, respectively, 8 and 9
are scan lines 10, 11, 1 containing apertures 6 and 7, respectively.
2 and 13 are end point candidate points, 14 is a vertical scanning line, 15 is a second region surrounded by two graphic strokes intersecting the vertical scanning direction, 16 is an opening of the second region 15, and 17
denotes a scanning line including the opening 16, 18 and 19 represent end point candidate points, and 20 and 21 represent overlapping regions where the first regions 4 and 5 overlap with the second region 15, respectively.

In the case of the present invention, as shown in FIG. 1A, the recognition target figure 1 is scanned, for example, in the horizontal direction, and
9. Set end point candidate points 10, 11, 12, and 13 at the intermediate positions of the respective widths of the two figure strokes of 9 soil, (4)
Next, as shown in FIG. 1B, the target figure 1 to be recognized is scanned in the vertical direction, and end point candidate points 18 and 19 are set at positions midway between the respective widths of the two figure strokes on the vertical scanning line 17. .

(111) Then, the set end point candidate point 10,1
Among the candidate points 1, 12, 13, 18, and 19, as shown in FIG. Points 10 and 19 are detected as end points. 0V)
In this case, as will be described later in FIG. 4, if necessary, the end point candidate point 10,
19 is detected as an end point.

FIGS. 2A, 2B, 2C, and 2C show explanatory diagrams illustrating an embodiment of the end point detection method according to the present invention.

In the figure, 1, 2, 3, 4, 5, 8, 9, 10, 11, 12
, 13, 14, 15, 17, 18, 19, 20, 21 respectively correspond to the symbols in FIG. 1, and 22, 23
, 24, 25 are singular line segments extracted by horizontal scanning, and 26, 27, 28, 29 are singular line segments 22, 23, 29, respectively.
24 and 25, 30 is a scanning line that touches the first area 4 on the upper side, 31 is a scanning line that touches the first area 5 on the lower side, and 32 and 33 are extracted by vertical scanning, respectively. 34 and 35 are points that are one ends of the singular line segments 32 and 33, respectively, 36 is a scanning line that touches the second area 15 on the right side, and 37 is a point that is one scanning distance after the horizontal scanning line 30. scan line,
Reference numerals 38 each represent a scanning line one scanning distance before the horizontal scanning line 8. In the case of this embodiment, as described later, (1)
The figure to be recognized 1 is scanned in the horizontal direction to extract singular line segments 22 and 23 that connect the intermediate positions of the two figure strokes surrounding the first area 4, and also to extract the singular line segments 22 and 23 that connect the intermediate positions of the two figure strokes surrounding the first area 4. 5, singular line segments 24 and 25 are extracted that connect the intermediate positions of each graphic stroke. (4) Then, end point candidate points 10, 11, Set 12 and 13. (111) On the other hand, the recognition target figure 1 is scanned in the vertical direction, and the singular line segment 3 connects the intermediate position of each figure stroke in two figure strokes surrounding the second area 15.
Extract 2,33. 0V) and points 18, 19, 34 located at both ends of each singular line segment 32, 33 extracted above.
, 35, end point candidate points 18 and 19 are set.

() Among the end point candidate points 10, 11, 12, 13, 18, and 19 set by these horizontal scans and vertical scans, points 11, 12, 13, and 18 are located in the overlap area 2.
0 and 21, so that they are not regarded as end points, and the remaining points 10 and 19 are detected as end points. The extraction of singular line segments and the setting of endpoint candidate points will be explained below with reference to FIG.

In FIG. 3, numeral 43 detects a black-white transition point (a transition point from a white area or a figure frame to a black area) from the inputted scanning information, and outputs, for example, one pulse signal at the time of detection. The black-and-white changing point determination circuit section 44 detects a black-white changing point (a transition point from a black area to a white area or a figure frame) from the input scanning information, and outputs, for example, one pulse signal at the time of detection. black and white change point determination circuits, 45 and 46 are comparison circuits, 47, 48, 49, respectively;
50 are counters, 51 and 52 are adder circuits, 53
, 54 are registers, 55 and 55' are not gates, 56 and 561 are counters, 57 and 5γ are flip-flop circuits, and 58 is an exclusive OR gate. For example, during one horizontal scan of the figure 1 to be recognized shown in FIG. A %clock signal %CL having a frequency of % is provided.

The counter 47 starts counting from the time when the scanning point reaches the x mark position shown in FIG. Counting is stopped when the time point b is reached. Therefore, when the counting is finished, the contents of the counter 47 represent the coordinate information of the point b. Similarly, the counter 48 has the value (c-b)/2 (c is A in FIG. 2).
Illustrated point. coordinate information). Further, the contents of the counter 49 indicate the coordinate information of the illustrated point d in FIG. 2A, and the contents of the counter 50 indicate the value (e-d)/2 (e is the coordinate information of the illustrated point e in FIG. 2A). As a result, the output of the adder circuit 51 is shown in FIG.
The output of the adding circuit 52 is the intermediate point f between the points d and e shown in FIG. 2A.
It will show the coordinate information of. The output of adder circuit 51 is set in register 53, and the output of adder circuit 52 is set in register 54. Note that the register 53
, 54, the contents based on scanning in which the scanning line does not intersect or contact the first regions 4, 5 are not gates 5.
5, 55'. In this way, during horizontal scanning, the singular line segments 22 and 24 shown in FIG.
Coordinate information of point a for each scan in and singular line segment 2
The coordinate information of point f for each scan in 3 and 25 is stored in register 53 and register 54, respectively. On the other hand, during vertical scanning, the same operation is performed, and the register 53 stores the point a'' for each scanning on the singular line segment 32 shown in FIG.
The coordinate information of the point f''(A' in FIG. 2) for each scan on the singular line segment 33 shown in FIG. 2 A'' is set in the register 54. be done.
Note that the above points a' and f'' are respectively point b shown in Figure 2 A''.
This is the midpoint between ``and c'' and the midpoint between points d'' and e'' shown in N in FIG. For example, the circuit shown in the lower part of FIG. 3 is used to set the end point candidate points, and the operation of this circuit will be explained. When using this circuit, the following processing is performed in advance, and the circuit is applied on that soil. That is, for example, during horizontal scanning, a process is performed in which a white area sandwiched between two graphic strokes, as shown by the hatched area in FIG. 2A, is pseudo-replaced with a black mesh point. The counters 56 and 56' in FIG. 3 convert the white area surrounded by the two figure strokes intersecting the scanning line into the black area (as described above).
The scanning point information obtained when scanning the recognition target figure formed by converting it into a character area) is used manually.

The counters 56 and 56'' are configured to count black mesh point information among the scanning point information.
The scanning point information inputted manually and the scanning point information inputted to the counter 56' correspond to scanning lines with a difference of one scanning. Counters 56 and 56'' count black mesh point information for each scan, and when the content reaches a preset threshold [α], the flip-flop circuit 5
7,57' is set. For example, in the case of horizontal scanning, scanning point information based on the scanning line 30 passing through the point 26 of the singular line segment 22 and the point 27 of the singular line segment 23 shown in FIG. scanning line 30
When scanning point information based on the scanning line 37 one scanning period after the flip-flop circuit 57 is input to the counter 567, the contents of the counter 56 and the counter 56'' are both equal to or higher than the threshold value "α" and the flip-flop circuit 57 and 57' are set together.

In this case, the output of exclusive-OR gate 58 is a logic "0"
Therefore, the points 26 and 27 are not considered as endpoint candidate points. Further, for example, scanning point information based on the scanning line 8 passing through the point 10 of the singular line segment 22 and the point 11 of the singular line segment 23 shown in FIG. When scan point information based on the scan line 38 one scan later is inputted to the counter 56'', the content of the counter 56 exceeds the threshold value "α", causing the flip-flop circuit 57 to be set. However, the content of the counter 56'' becomes less than the threshold value "α" and the flip-flop circuit 5r is maintained in the reset state. Therefore, the output of the exclusive OR gate 58 becomes a logic "1", and therefore the points 10 and 11 are set as endpoint candidate points. Below, the second
Twelve illustrated points 13, 18, and 19 are set as end point candidate points, respectively. Although the end point candidate points 10, 11, 12, 13, 18, and 19 set in this way are not described in detail, by checking whether or not they correspond to the overlap areas 20 and 21 as described above, the overlap areas 20 and 21 are determined. End point candidate points 11, 12, 13 corresponding to regions 20, 21
, 18 are not considered as endpoints, and the remaining endpoint candidate points 10,
19 is detected as an end point.

FIG. 4 A, B, and C show the end points 10 and 10 detected in this way.
19 is an explanatory diagram for explaining the process of checking the inclination angle of a graphic stroke extending from 19.

In carrying out this process, as shown in FIG.
, 7 and a line segment including the opening 16 of the second region 15 marked with ±.

In the illustrated areas α and β in FIG. 4A, points 59 and 6 where the graphic stroke intersects or touches the line segment constituting the graphic frame 2 or the areas α and β from the end points 10 and 19, for example.
0, 61, and 62, graphic strokes are observed in the directions of arrows A and B shown in the figure, and a histogram is created. FIG. 4B shows the types of directional strokes in, for example, a 3×3 mesh observation window used when performing this observation.

Moreover, FIG. 4C shows a histogram created by the above-mentioned observation. As is clear from FIG. 4C, as a result of observing the end point 10 in the direction of the arrow A shown in FIG. 4 LA, the directional stroke shown in FIG. The results observed in the fourth
The directional strokes shown in Figure B are the most numerous.

On the other hand, regarding the end point 19, the number of directional strokes is the greatest in the direction of the arrow A shown in FIG. 4A, and the number of directional strokes is the most in the direction of the arrow B. That is, end point 1
The graphic strokes extending from 0 and 19 are arrow A.
It can be seen that the directional strokes are different in the direction and in the direction of arrow B. Therefore, it is confirmed that end points 10 and 19 are indeed end points. As described above, according to the present invention, when recognizing a figure to be recognized, it is possible to accurately detect end points using the original figure information as is without performing a thinning process.

[Brief explanation of the drawing]

FIGS. 1A, B, and C are explanatory views for explaining aspects of the endpoint detection method according to the present invention, and FIGS. 2A, B, A'', W, and C are explanatory diagrams for explaining an embodiment of the endpoint detection method according to the invention Figure 3 shows an embodiment of the configuration of the main part of the present invention for realizing an embodiment of the end point detection method shown in Fig. 2, and Fig. 4 A, B, and C show graphic strokes extending from the end point candidate point. An explanatory diagram for explaining the process of checking the inclination angle of 1, openings in areas 4 and 5, 10, 11, 12, and 13, end point candidate points, 14, vertical scanning line, 15, second area, 16, openings in second area 15, 18, 19, Endpoint candidate points 20 and 21 represent overlapping regions, respectively.

Claims (1)

[Scope of Claims] 1. In a pattern recognition device that performs recognition processing of the recognition target figure by extracting features based on scanning information obtained by scanning the recognition target figure, the recognition target figure is scanned in the horizontal direction. a singular line segment extraction unit that extracts a singular line segment during horizontal scanning by determining a point at an intermediate position of the width of each of the two graphic strokes that intersects with the horizontal scanning line; An end point candidate point extraction setting unit that detects a first area surrounded by two figure strokes that intersect with and extracts both end points of the singular line segment during horizontal scanning as end point candidate points; a singular line segment extraction unit that extracts a singular line segment during vertical scanning by determining a point in the middle of the width of each of the two graphic strokes that intersects the vertical scanning line when scanning;
an end point candidate point extraction setting unit configured to detect a second area surrounded by two figure strokes intersecting the vertical scanning line and extract both end points of the singular line segment during vertical scanning as end point candidate points; Among the extracted and set end point candidate points, the end point candidate points corresponding to the overlap region where the first region and the second region overlap are not regarded as end points, and the remaining end point candidate points are detected as end points. An end point detection method characterized by the following. 2. The end point detection according to claim 1, wherein the remaining end point candidate points are subjected to a process of checking the inclination angle of a graphic stroke extending from the end point candidate points, and are detected as end points. method.