JP2849101B2 - Loop determination method for line figures in character recognition - Google Patents

Description

The present invention relates to a method of determining a loop of a line figure for recognizing a figure pattern obtained by thinning a figure such as a handwritten character. .

(Prior Art) Japanese Patent Application Laid-Open No. Sho 62-16 discloses a technique for determining the inclusion relation of a closed figure.
There is one disclosed in Japanese Patent No. 0584. This is the sixth publication of the publication.
When there are two closed figures A and B as shown in FIGS. 9A and 9B, the inclusion relation between these two closed figures is determined. In FIG. 6A, A and B do not have an inclusive relation to each other, and B is included in A in FIG.

(Problems to be Solved by the Invention) However, the inclusion relation described in this publication is such that one closed figure completely encloses another closed figure (no common line segment) or is completely separate. It is just judging whether it is. Therefore, with this determination technique, it is not possible to determine the inclusion relationship when a part of the line segment of the two closed graphics is common. Because, in this determination technique, when two closed figures do not intersect or touch, they are once traced, and then the two figures are virtually connected by a line segment for tracking, so that both figures share the line segment from the beginning. This is because the tracking of the curve itself is not determined before connecting the virtual line.

Therefore, an object of the present invention is to determine the inclusion relationship between two closed figures sharing a part of a line segment and to separate the two into two closed figures having no inclusion relationship when there is an inclusion relation. It is an object of the present invention to provide a method for determining a loop of a line segment figure.

SUMMARY OF THE INVENTION (Means for Solving the Problems) The present invention relates to a loop determining method in character recognition, and an object of the present invention is to provide a method for thinning a figure such as a character when recognizing the figure as an end point. Alternatively, a node at a branch point and a branch of a line segment connecting the node are expressed as a loop, and a closed curve formed by the branch is used as a loop, and the node and the branch are tracked to obtain tracking data. The number of detected loops is defined as a first loop and a second loop, and all the branches constituting the first loop and the second loop are used in duplicate with the two loops of the first loop and the second loop. Third
A first branch group obtained by deleting the third branch group from the first loop; and a second branch group obtained by deleting the third branch group from the second loop. The three types of branch groups are developed on xy coordinates with a common point as K, and the developed three types of branch groups are scanned in the xy direction. In this scan, two types of the branch groups are found. At this point, the scanning is stopped, the two types of branch groups are determined as outer branch groups, and the remaining group is defined as a common branch group. The two types of outer branch groups and the common branch group are used as the two loops. This is achieved by reconstructing

(Operation) In the character recognition of the present invention, tracking of a line graphic such as a thinned character is performed as follows. That is, the starting point is determined at the intersection of the scanning lines or any one node, the first tracking is performed, and after the branch is tracked to the next node,
After registering whether the reached node has an untracked branch or not, return to the original starting point and determine whether or not the starting point still has an untracked branch, and there is an untracked branch. In this case, the tracing is performed to the next node of the branch in the order according to the predetermined tracing search order, and when the node is reached, the untracked or tracked branch is registered here, and then the first node of the branch is performed. The same tracking is sequentially performed on the untracked branch at the returned node in the order according to the predetermined tracking search order, and each time the node is reached, the process returns to the original node and the untracked branch is returned. It is determined whether or not there is. Repeat the above tracking operation until there are no untracked branches, and when there are no untracked branches, go to the node reached in the first tracking and refer to the registered data, and The operation of performing tracking to the next node according to the direction of the predetermined tracking search order is repeated. As a result of performing such tracking, even in the case where a plurality of closed loops are configured with elements of a common line segment in a closed loop configuration for a line figure, when a plurality of closed loops are recognized, they have an inclusive relation as a figure. Separation processing so that it does not
Since the loop is reconstructed by expanding and examining the positional relationship between the outside and inside of the branch group on the coordinates, more accurate character recognition can be realized.

(Embodiment) FIG. 1 is a diagram for explaining a method of tracking a thinned figure which is a premise of the present invention.
The figure shows an example of numbers (continuous writing of 2, 3) composed of branches BR1 to BR10 between N1 to N10 and each node, and the scanning for tracking is a scanning line SL that moves in the y direction and the x direction. An example is shown below. In the character line figure in this case, since the node N1 intersects the scanning line SL first, the node N1
Traces branch BR1 first, starting at node N2
, Once it returns to the first starting point N1, this starting point N1
To determine if there is an untracked branch. In this case, since there is no untracked branch, the process proceeds to the node N2 obtained in the previous tracking to determine whether or not there is an untracked branch at this node N2. In this example, the branches BR2 and B
Since R3 has not been tracked, the branch BR2 having a small number according to the tracking search order “1” → “8” as shown in FIG.
Is tracked first. The determination of the tracking search order is performed by a mask pattern as shown in FIG. 2, the center of the mask is adjusted to the target dot, and the tracking search order is determined by the direction codes of the corresponding eight peripheral pixels. .

When reaching node N3, return to node N2, this node
N2 determines if there are any untracked branches yet,
In this example, since the branch BR3 has not been tracked, the branch BR3 is tracked and finally reaches the node N5. When it reaches this node N5, it returns to node N2 again, and this node N2
It is determined whether or not there is an untracked branch. In this example, since there is no untracked branch, the process first proceeds to the node N3, and from this node N3 in the direction according to the tracking search order in FIG. The untracked branch (the loop indicated by BR4 in this example) is tracked, and it is determined whether or not there is an untracked branch at the node N3 when the node N3 reaches the start point. In this example, since the branch BR5 has not been tracked, the branch BR5 tracks the node N4, and then returns to the node N3. Node N3
Then, the process proceeds to the node N5, which is the next destination obtained by tracking the node N2, and it is determined whether or not there is an untracked branch at this node N5. In this example, the branch BR6 according to the tracking search order is first traced to the node N6, and the process returns to the node N5 to trace the next branch BR7 to the node N7. To node N8. When the tracing is performed to the node N8, the process returns to the node N5 again, and it is determined whether or not there is an untracked branch at the node N5. In this example, since all the tracing has been completed, the process proceeds to the node N8. Then, the node N8 determines the untracked branch in accordance with the tracking search order and then performs tracking. And then return to node N8 again. Returning to the node N8, at this point, there is no untracked branch, and there is no untracked node that is not an end point, so it is known that the tracking has been completed.

In the above-described line figure tracing, when a single branch is traced and its end node is a branch point, and the end node is registered before tracing the current branch, a loop has occurred. become. For the tree structure of the tracking data, for example, numerical names are given to nodes and branches as shown in FIG. That is, the number is the node name, and the underlined number is the branch name. In FIG. 4, the loop occurred at node 1112.
At the time when the tracking of the branch 11122 is completed. In other words, this is the point in time when it is found that the nodes 1121 and 11122 are at the same point. Therefore, two names are given to the node A, and the constituent branch is identified from the names. That is, the starting point is the node 11 from the part having the same name, and the constituent branch name is It can be seen that the number of constituent branches is “5”. Next, as post-processing, the name for the node A is changed to a name given later. This means that the starting point is updated, the number of branches forming the loop can be reduced as much as possible, and the loop itself can be reduced. FIG. 5 shows an example of this. When it is determined that the nodes 121 and 1121 are the same, the first loop is determined. The starting point is 1, and the constituent branches are 12 , 121 , 11 , 112 , 1121 . When it is determined that the nodes 1122 and 11211 are the same, the second loop is determined. The starting point is 112, and the constituent branches are 1122 , 1121 , 11211 . In FIG. 5, if the post-processing of the first loop is not performed, the node [11211] in FIG. 5 becomes 1211 (not shown). In this case, when it is determined that the nodes 1211 and 1122 are the same, the second loop is determined, and the starting point is 1,
The constituent branches are 12 , 121 , 1211 (not shown in FIG. 5, but the 11211 branch), 11 , 112 , 1122 , and the loop becomes large.

Next, a case of a loop composed of only one branch will be described. As shown in FIG. 6, when the second loop is composed of one branch, no post-processing is performed. This is because there is no need to update the starting point. In the example of FIG. 7, there is no need for post-processing.

Whether or not the loop is composed of one branch can be determined based on whether or not the node whose end point of the tracked branch coincides is the start point of the branch. As a specific method, information indicating a parent branch when the node is registered is provided for a node, and information indicating a node at the start point is provided for a branch. By using this information to search from the node where the loop has occurred to the start point, the constituent branch can be extracted.

Next, a description will be given of a method of separating a plurality of loops in which branches are used redundantly so that there is no inclusive relation.

When a loop is created by the above-described method in the trace diagram of the graphic pattern as shown in FIG. 8A, two loops whose inclusion relation is not clear are created. At this time, the inclusion relation cannot be expressed only by the tracking order. Considering two adjacent loops, that is, a loop in which branches are used in duplicate, the tracking diagram in FIG. 8B shows two nodes A and B as shown in FIG.
And three types of branch groups (here, tentatively one group… 1,4,2
Groups: 3, 9, 11, 3 groups: 2, 7).
FIG. 8C shows the branch names represented by underlined numbers. The correspondence is 11 for 1, 12 for 2, 13 for 3,
111 is 4, 112 is 5, 121 is 6, 122 is 7, 131 is 8, 132 is 9, 1321 is 10, and 1322 is 11. FIG. 8D shows a tree structure of the tracking data. From this tree, the first loop (Loop 1) is 11 , 12 , 111 , 121 , ie, 1,2,4,7
And the second loop (Loop 2) is 12 , 13 , 122 , 132 ,
1322 , that is, 2,3,7,9,11. And
In order to have no inclusive relation, which of the three types of branch groups is the contact line of two loops (common part)
It is sufficient to determine whether

(I) Step 1: For two loops obtained from the tracking result, the branches from the constituent branches are classified into three types of branch groups. That is, the constituent branch of the loop 1 shown in FIG. 10 (A) is “1,2,4,7”, and the constituent branch of the loop 2 shown in FIG. 10 (B) is “2,3,7,9, 11 "and two loops 1
If a plurality of (and sometimes one) overlapping branches of the second and third branches are defined as a third branch group, the third branch group = 2,7. A branch group obtained by removing the third branch group from the constituent branches of the loop 1 is referred to as a first branch group.
Assuming that a branch group obtained by removing the third branch group from the constituent branches of the loop 2 is a second branch group, the first branch group to 1,4 and the second branch group = 3,9,11. (See FIG. 10 (C)).

(Ii) Step 2: The three branch groups obtained in the above step 1 are
Expand on the xy coordinates as shown in the figure. At the time of expansion, the name of each branch group (“1”, “2”, “3” in this example) is given, and the common point is “K”.

(Iii) Step 3-1: On the developed coordinates, branch groups 1 to 3 as shown in FIG.
And scanning from the outside until the common point K is reached. In this case, the main scanning is in the direction A and the sub-scanning is in the direction B.
Perform in the order of (1) → (2) → (3) → (4) in FIG. Note that the main scanning and sub-scanning directions may be reversed, and (1)
The scanning order of (4) can be arbitrarily changed. Scanning is stopped when two of the first to third branch groups are found at the time of the scan, the two sets are determined as the outer branch groups, and the remaining group is set as the common branch group. In the example shown in FIG. 12, the eighth scan from the top in the direction (1) in FIG. 11, that is, the second scan in the sub-scanning B direction is the eighth scan and the second branch group is found in the fourth scan from the left. At this point, the third and first branch groups are determined to be outside, and scanning is terminated.

(Iv) Step 3-2: If only one branch group is found in step 3-1 above, the line of the found group is scanned from the common point K, and a point including the outer region near 8 is searched. From this point, an outer area is sequentially determined in the vicinity of 8 to the common point K. After the completion of the common point K, each end point of the other two branch groups that are not in the vicinity of 8 in the common point K is included in the 8 neighborhoods. Check for outside areas. In this case, one of the points is always in contact with the outer region, and the group in contact with the outer region ends as the second outer branch group. That is, FIG. 13 (A)
The branch group of the graphic pattern as shown by (2) is as shown in FIG. 3 (B), and only one branch group is found in step 3-1. 13 of the tracking diagram in FIG. 13 (B)
Pieces of one of the 14 views of as K _a common point K, K _b and the other (the both variable as K _a, K _b) and then, eight neighbor shown in FIG. 15 around the K _a point (A) , It is determined whether or not a portion of the outer region is included. The outer region here is the outer portion scanned in step 3-1 described above, that is, the hatched portion in FIG. In this case, the outer region so not to 8 near the K _a point, continued scanning to a point where an outer area 8 near along the group of a line found at step 3-1. That is, because one step 3-1 are found, K _a
Next to the point, as shown in FIG. 15 (B), the vicinity of point 8 of the upper right “1” is seen. However, since it does not include the outer region, the point of the upper right “1” is further shown in FIG. 15 (C). Look around 8 of. As a result, since the outer region of the hatched portion is found, the outer region is extended to the pixel EN1 in the eight neighboring regions as shown in FIG. 15 (D). When repeated until successively K _a point above extended operation, made as a FIG. 15 (E) in FIG. (F), not that it can for other Regions in 8 near the K _a point. Note that points EN2 and EN3 in FIG. 15E indicate expanded pixels. Then, returning to the K _a point for each end point of the other two groups within 8 near the K _a point, to check whether the outer region present in the 8-neighborhood. In this case, as shown in FIG. 15 (G), there is no outside area near 8 of the end point of “2” connected to the common point K ₂ , but as shown in FIG. Since there is an outer region near the vicinity 8, the remaining one group in contact with the outer region is determined as the second branch group.

(V) Step 4: Two loops are reconstructed so that the inside is a common branch group by the two branch groups that are boundaries with the outside region and the remaining inside group. In the case of the graphic pattern shown in FIG. 9, the first loop and the second loop shown in FIGS. 16A and 16B are reconstructed. In the case of the graphic turn shown in FIG. The first loop and the second loop as shown in FIGS. 17A and 17B are reconfigured.

In steps 1 to 4 described above, the inclusion relationship between two loops having a common branch group can be checked, and loops that are regionally independent can be reconstructed.

Effects of the Invention As described above, according to the loop determination method of the present invention, when a plurality of closed loops are configured by common line segments in a closed loop configuration, these closed loops are included as graphics. Since the data is separated so as not to have a relationship and the outer and inner relationships are examined and reconstructed, more accurate character recognition can be realized.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining a tracking principle which is a principle of the present invention, FIG. 2 is a diagram showing an example of a mask pattern used for tracking, and FIG. 3 is a diagram for explaining an example of a direction of a tracking search order. FIGS. 4 to 7 are diagrams for explaining creation of a closed loop, FIG. 8 is a diagram showing an example of figure tracking, and FIGS. 9 to 17 (A) and (B) are the inventions, respectively. FIG. 4 is a diagram for explaining loop reconfiguration of FIG. N1 to N10 ... nodes. BR1 ~ BR10 ... branch, n1 ~ n4, N
10 to n13: Node, br1 to br3, br10 to br12 ... Branch.

Continuation of the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06K 9/46 G06T 7/00

Claims (2)

(57) [Claims] When a figure such as a character is thinned and recognized, the line figure is represented by a node at an end point or a branch point and a branch of a line connecting the node, and a closed curve formed by the branch is represented. A loop is determined by tracing the node and the branch to obtain tracking data. When two loops are detected based on the tracking data, a first loop and a second loop are determined.
A first branch and a second loop, wherein the first and second loops are all divided into a third branch group which is used by overlapping with the first and second loops;
A first branch group in which the third branch group is deleted from the loop; and a second branch group in which the third branch group is deleted from the second loop, the three types of branch groups. Are developed on the xy coordinate with the common point as K, and the developed three types of branch groups are scanned in the xy direction. In this scanning, two of the branch groups are scanned.
When the type is determined, the scanning is stopped, the two types of the branch groups are determined as the outer branch groups, and the remaining group is defined as the common branch group. A loop determining method for a line figure in character recognition, wherein loops are reconstructed. 2. When only one of the branch groups is found in the scanning in the xy direction, the line of the found one branch group is scanned from the point K to search for a point including the outer region near eight. Then, the outer region is determined in the vicinity of the eight points from the search point to the point K in sequence, and after the completion of the point K, the other two groups within the vicinity of the point K are subjected to the determination of the vicinity of the eight points. 2. The method according to claim 1, wherein it is determined whether or not the outside area exists in the inside, and a group in contact with the outside area ends as a second outside branch group.