Recognition method for letter patterns

• US 4,987,603 A
• Filed: 09/22/1988
• Issued: 01/22/1991
• Est. Priority Date: 09/22/1988
• Status: Expired due to Term

• Current Assignee: Glory Kogyo Kabushiki Kaisha
• Original Assignee: Glory Kogyo Kabushiki Kaisha
• Inventors: Fujita, Yasuo, Ohnishi, Kazuhiko, Uosaki, Katsuji, Miki, Shoji, Tsubota, Hiroki
• Primary Examiner(s): Moore, David K.
• Assistant Examiner(s): COUSO, YON JUNG
• Application Number:
  US07/247,664
• Time in Patent Office: 852 Days
• Field of Search: 382/25, 382/26, 382/13, 382/21, 382/38, 382/34
• US Class Current: 382/203
• CPC Class Codes: G06V 10/44 Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/2504 Coarse or fine approaches, e.g. resolution of ambiguities or multiscale approaches

Abstract

A letter pattern recognition method preprocesses each character pattern to effect characteristic extraction based on elements such as a loop, a straight line and an arc and is adapted to compare the character pattern with a preregistered classification code after the normalization thereof, the character pattern being subjected to balance and stroke checks if necessary as a result of the comparison. Notwithstanding this, if the character is not defined, the character pattern is identified by means of detail classifications such as those of a divergent mode of the arc, and a length ratio of the elements and the like.

Claims

1. A letter pattern recognition method comprising the steps of:

preprocessing character patterns;

extracting characteristic data for the character patterns according to elements including a loop, a straight line and an arc which have branches and nodes;

obtaining a center of gravity for each of the respective elements;

normalizing the character patterns by using the extracted characteristic data;

comparing the normalized character patterns with pre-registered classification codes by allotting relative positions of the loop, straight line and arc elements in accordance with information on the center of gravity thereof and determining if a return check is required depending upon the result of the comparison, and performing a return check and returning to the extracting step if a return check is required;

performing balance and stroke checking for the normalized patterns in the absence of the requirement for the return check, wherein the balance checking is performed by using a relationship of weight on the relative position and wherein the stroke checking is performed by checking a direction of the straight line elements, a rotational direction of the arc elements and a link structure of the elements; and

detail-classifying the normalized patterns in such a manner that an orientation of the arc elements, and a length ratio of the elements so as to thereby recognize the character patterns.

2. A letter pattern recognition method as set forth in claim 1, wherein a code with noise and another code without noise are prepared in advance for the classification codes, and wherein when the code with noise is anticipated in establishing the comparison, the comparison is made again after eliminating elements corresponding to unessential elements defined in advance in the code with noise.

3. A letter pattern recognition method as set forth in claim 1, wherein the character patterns are handwritten figures.

4. A letter pattern recognition method as set forth in claim 3, wherein the classification codes are codes including "0"˜

"9".

5. A letter pattern recognition method as set forth in claim 1, wherein the loop element is extracted by a linear pattern tracing algorithm and refers to a pattern which defines a loop selected from the group of loops consisting of a closed loop and a ring shape loop, said loop being formed by at least one branch.

6. A letter pattern recognition method as set forth in claim 1, wherein there is provided a branch which is formed of two points, and a segment of which is of a length which is the shortest distance md between points A and B, and wherein the branch Ll is recognized as a straight line if the branch tl meets the following inequality:

space="preserve" listing-type="equation">α

.md≧

tl

where α

;

straight line determining coefficient (for example 1.2) ##EQU11## C_n ;

number of chain code Cl_n ;

1(chain code stands at 1, 3, 5 or

     7)

√

2 (chain code stands at 2, 4, 6 or

     8).

7. A letter pattern recognition method as set forth in claim 1, wherein the chain codes are designated 1˜

8 with respect to directions of the straight line.

8. A letter pattern recognition method as set forth in claim 1, wherein the arc element is a characteristic element other than those of the loop and straight line, and has characteristic codes corresponding to divergent orientation and divergent amount.

9. A letter pattern recognition method as set forth in claim 8, wherein if a point A of one end point of the arc element and another point B of another end point thereof are respectively expressed by (xA, yA) and (xB, yB), and a median point M has the coordinates which are formularized by:

##EQU12## and the direction of a segment that connects a point CP with the point M is expressed by;

##EQU13## the point CP is a median point of the arc element whose end points are A and B

wherein d=yd/xd is given, and then a divergent orientation n of the arc is formularized by ##STR2##

10. A letter pattern recognition method as set forth in claim 1, wherein the branches other than those that form the loop elements are employed as single branches, and wherein output data including all data regions, respective numbers of loops, straight lines and arcs, and a ratio for the loop, straight line and arc elements are stored when the single branches are completely processed.

11. A letter pattern recognition method as set forth in claim 10, wherein a determination as to whether or not a segment is a straight line is made when a single branch has not yet been processed to a full extent, and wherein the single branch is subjected to a straight line process for processing the straight line when it is determined to be a straight line, and wherein a determination is made as to whether or not the single branches number more than two, and wherein the single branches are subjected to an arc process for processing the arc when the single branches number more than two.

12. A letter pattern recognition method as set forth in claim 11, wherein when the number of single branches is one, arc and loop flags for processing the arc and loop are respectively set equal to ON and OFF.

13. A letter pattern recognition method as set forth in claim 12, wherein the branch is subjected to a separation process for separating the branch when neither the arc nor loop flags have been set equal to OFF.

14. A letter pattern recognition method as set forth in claim 13, wherein the separation process is a process that separates a pattern formed by only one branch and is not a straight line into two elements.

15. A letter pattern recognition method as set forth in claim 14, wherein a determination as to whether or not one separated element of the two separated elements is a straight line is initially made after the separation process, and wherein the separated element is subjected to the straight line process when the element is determined to be straight line, and wherein the separated element is subjected to the arc process when it is determined to be non-rectilinear, and the other separated element of the two separated elements is then subjected to the same arc process.