Pattern recognition apparatus

• US 4,783,829 A
• Filed: 02/22/1984
• Issued: 11/08/1988
• Est. Priority Date: 02/23/1983
• Status: Expired due to Term

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Hata, Seiji, Miyakawa, Akira, Nishida, Yoshie
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): Mancuso, Joseph
• Application Number:
  US06/582,555
• Time in Patent Office: 1,721 Days
• Field of Search: 382/22, 382/16, 382/25, 382/30, 382/34, 382/41, 382/44, 382/56, 358/260
• US Class Current: 382/199
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06V 20/66 Trinkets, e.g. shirt buttons or jewellery items recognising microscopic objects G06V20/69

Abstract

A pattern recognition apparatus for recognizing a pattern derived from image data of an outline of an object region is disclosed. The apparatus comprises means for identifying the object region from the input image data, means for extracting outline pixels of the object region, means for polygon-approximating the object region based on a position relationship among the outline pixels to determine vertex coordinates of the approximation polygon, means for determining characteristic data representing a position and an attitude of the region based on the vertex coordinates, and means for superimposing the polygon-approximated object region pattern on a dictionary pattern based on the characteristic data and, determining a common region of the patterns to determine the identity of the patterns. High speed and highly reliable pattern matching is attained.

Claims

1. A pattern recognition apparatus, comprising:

(a) means for identifying an object region of an object from input image data;

(b) means for extracting outline pixels of the identified object region;

(c) means for polygonally-approximating an outline of said object region based on a positional relationship of the extracted outline pixels and for determining vertex coordinates of a polygonal approximation of the outline of the identified object region;

said vertex coordinate determination means including;

a polygonal approximation data memory for storing pixel pattern data which can be approximated as lines of a polygonal approximation of the outline from the input image data;

means for determining end points of segments of the outline by referring to the pixel pattern data stored in said polygonal approximation data memory based on connectivity among the extracted outline pixels, and for sequentially segmenting said outline into the segments; and

a vertex data memory for storing data about the end points of the segments to provide end point data as the polygonally-approximated vertex coordinates;

(d) means for determining characteristic data representing the center and the orientation of said polygonally-approximated region based on said vertex coordinates; and

(e) means for superimposing a polygonally-approximated region pattern on a dictionary pattern based on said characteristic data to generate a common region by matching the center of said polygonally-approximated region pattern with the center of said dictionary pattern and transforming the vertex coordinates of said polygonally-aproximated region pattern on the basis of any angular difference between the orientation of the polygonally approximated region pattern and orientation of said dictionary pattern to provide a transformed region pattern, and for determining a degree of matching between the transformed region pattern and the dictionary pattern.

2. A pattern recognition apparatus according to claim 1 wherein said identifying means (a) includes:

input control means for binarizing said input image data by predetermined threshold;

an image memory for storing said binarized image data;

processing means for identifying the object region based on the image data stored in said image memory and simultaneously determining position information of a representative point of said region; and

a region information memory for storing said position information of said representative point.

3. A pattern recognition apparatus according to claim 2, wherein said outline pixel extracting means includes:

outline extracting means for determining a start point of scan based on said position information of said representative point, sequentially scanning the outline pixels of the object region on the image and determining center coordinates of said outline pixels; and

an outline memory for storing said center coordinates.

4. A pattern recognition apparatus according to claim 3, wherein said vertex coordinate determination means includes:

polygonal approximation means for polygonally-approximating the sequence of said center coordinates within a predetermined allowable error range, to polygonally-approximate said object region and sequentially determine the vertex coordinates; and

a polygonal approximation information memory for storing said vertex coordinates.

5. A pattern recognition apparatus according to claim 4, wherein said characteristic data determination means includes:

region characteristic extraction means for extracting region characteristic data including said center and orientation of said polygonally-approximated object region from the content of said polygonal-approximation information memory; and

a region characteristic memory for storing said region characteristic data.

6. A pattern recognition apparatus according to claim 5, wherein said matching determination means includes;

common region extraction means for superimposing said object region pattern on said dictionary pattern by transforming the coordinates thereof based on the center and orientation data of said polygonally-approximated region in said characteristic data to determine a common region of said patterns as a sequence of vertex coordinates;

a common region memory for storing the sequence of vertex coordinates of said common region and

recognition means for calculating an area of said common region based on the vertex coordinates stored in said common region memory, calculating a ratio of said common region area to an area of one of said patterns, and determining the matching of said patterns when said ratio is larger than a predetermined value.

7. A patten recognition apparatus according to claim 3, wherein said vertex coordinate determination means includes:

a polygonal approximation data memory for storing polygonal approximation data for a point pattern to polygonally-approximate the outline pixels of the input image data;

means for determining end points of segments by referring to the polygonal approximation data stored in said polygonal approximation data memory based on connectivity among the extracted outline pixels, and sequentially segmenting said outline pixels; and

a vertex data memory for storing data about the end points of the polygonal approximation to provide said end point data as the polygonally-approximated vertex data.

8. A pattern recognition apparatus according to claim 1 wherein said outline pixel extracting means (b) includes:

outline extracting means for determining a start point of scan based on position information of a representative point of said region, sequentially scanning the outline pixels of the object region on the image and determining center coordinates of said outline pixels; and

an outline memory for storing said center coordinates.

9. A pattern recognition apparatus according to claim 1 wherein said vertex coordinate determination means (c) includes:

polygonal approximation means for polygon-approximating the sequence of center coordinates of said outline pixels within a predetermined allowable error range, to polygonally-approximate said object region and sequentially determining the vertex coordinates; and

a polygonal approximation information memory for storing said vertex coordinates.

10. A pattern recognition appartus according to claim 1 wherein said characteristic data determination means (d) includes:

region characteristic extraction means for extracting regio characteristic data including said center and orientation of said polygonally-approximated object region from said vertex coordinates; and

a region characteristic memory for storing said region characteristic data.

11. A pattern recognition apparatus according to claim 1, wherein said matching determination means (e) includes:

common region extraction means for superimposing said object region pattern on said dictionary pattern to determine the common region between said transformed region and dictionary patterns as a sequence of vertex coordinates;

a common region memory for storing the sequence of vertex coordinates of said common region; and

recognition means for calculating an area of said common region based on the vertex coordinates stored in said common region memory, calculating a ratio of said common region area to an area of one of said patterns, and determining said degree of matching of said patterns when said ratio is larger than a predetermined value.

12. A pattern recognition apparatus according to claim 1 wherein said matching determination means (e) includes means for identifying the position and the attitude of the polygonally-approximately region from relative position information about said patterns.

13. A pattern recognition apparatus, comprising:

(a) preprocessing means for binarizing input image data, identifying an object region based on said data and concurrently determining position information of a representative point of said region;

(b) outline extraction means for sequentially searching outline pixels of said object region based on said position information of said representative point to extract the outline pixels;

(c) polygonal approximation means for polygonally-approximating an outline of said object region based on position relationship of said outline pixels by referring to a table of a pattern series capable of approximation of lines and storing vertex coordinates of the polygonally-approximated region in a memory;

(d) region characteristic extraction means for extracting region characteristic data about the center and the orientation of said polygonally-approximated region from said vertex coordinates and storing the region characteristic data in a memory; and

(e) pattern matching means for superimposing a polygonally-approximated pattern associated with the content stored by said polygonal approximation means and said region characteristic extraction means on a previously prepared dictionary pattern to generate a common region by matching the center of said polygonally-approximated region pattern with the center of said dictionary pattern and transforming the vertex coordinates of said polygonally-approximated region pattern on the basis of any angular difference between the orientation of the polygonally approximated region pattern and orientation of said dictionary pattern to provide a transformed region pattern, and extracting a common region common to both the transformed region pattern and the dictionary pattern, and dictionary patterns when an area of said common region is larger than a predetermined area.

14. A pattern recognition apparatus according to claim 13 wherein said preprocessing means (a) includes:

input control means for binarizing said input image data by a predetermined threshold;

an image memory for storing said binarized image data;

processing means for identifying the object region based on the image data stored in said image memory and simultaneously determining position information of a representative point of said region; and

a region information memory for storing said position information of said representative point.

15. A pattern recognition apparatus according to claim 14 wherein said outline extracting means (b) includes:

outline extracting means for determining a start point of scan based on said position information of said representative point, sequentially scanning the outline pixels of the object region on the image and determining center coordinates of said outline pixels; and

an outline memory for storing said center coordinates.

16. A pattern recognition apparatus according to claim 15 wherein said polygonal approximation means (c) includes:

polygonal approximation means for polygon-approximating the sequence of said center coordinates within a predetermined allowable error range, to polygon-approximate said object region and sequentially determining the vertex coordinates; and

a polygonal approximation information memory for storing said vertex coordinates.

17. A pattern recognition apparatus according to claim 16 wherin said region characteristic extraction means (d) includes:

region characteristic extraction means for extracting region characteristic data including said center and orientation of said polygonally-approximately object region from the content of said polygonal approximation information memory; and

a region characteristic memory for storing said region characteristic data.

18. A pattern recognition apparatus according to claim 13 wherein said pattern matching means (e) includes:

common region extraction means for superimposing said pattern on said dictionary pattern by transforming the coordinates thereof based on the center and the orientation data of said polygonally-approximated region in said characteristic data to determine the common region of said patterns as a sequence of vertex coordinates;

a common region memory for storing the sequence of vertex coordinates of said common region; and

recognition means for calculating an area of said common region based on the vertex coordinates stored in said common region memory, calculating a ratio of said common region area to an area of one of said patterns, and determining the matching of said patterns when said ratio is larger than a predetermined value.

19. A pattern recognition apparatus according to claim 13 wherein said pattern matching means (e) includes means for identifying the center and orientation of the object from relative position information of said patterns.

20. A pattern recognition apparatus according to claim 13 wherein said polygonal approximation means (c) includes:

a polygonal approximation data memory for storing polygonal approximation data for a point pattern required to polygonally-approximate the outline pixels of the digital image;

means for determining end points of segments by referring to the polygonal approximation data stored in said polygonal approximation data memory based on connectivity among the extracted outline pixels, and sequentially segmenting said outline pixels; and

a vertex data memory for storing the end point data of the polygonal approximation to output said end point data as the polygon-approximated vertex data.

21. A pattern recognition apparatus, comprising:

(a) preprocessing means for binarizing input image data, identifying an object region from said data and concurrently determining position information of a representative point of said region;

(b) outline extraction means for sequentially searching outline pixels of said object region based on said position information of said representative point and determining center coordinates of said outline pixels;

(c) polygonal approximation means for polygonally-approximating said object region based on said center coordinates by referring to a table of a pattern series capable of approximation of lines and storing vertex coordinates in a dictionary pattern memory when said input image data relates to a dictionary pattern and otherwise storing said vertex coordinates in a polygonal approximation information memory;

(d) region characteristic extraction means for extracting characteristic data about the center and the orientation of said polygonally-approximated object region from said vertex coordinates and storing the region characteristic data in said dictionary pattern memory when said input image data relates to said dictionary pattern and otherwise storing said region characteristic data in a region characteristic memory; and

(e) pattern matching means for positioning a polygonally-approximated pattern associated with the contents stored in said polygonal approximation information memory and the region characteristic memory and a polygonal dictionary pattern associated with the content of said dictionary pattern memory previously prepared based on said region characteristic data with respect to the center and the orientation to superimpose the polygonally-approximated and polygonal dictionary patterns to generate a common region by matching the center of said polygonally-approximated pattern with the center of said polygonal dictionary pattern and transforming the vertex coordinates of said polygonally-approximated pattern of the basis of any angular difference between the orientation of the polygonally approximated pattern and orientation of said polygonal dictionary pattern to provide a transformed region pattern and extracting a common region common to both the transformed region and the polygonal dictionary patterns when a ratio of an area of said common region to an area of said dictionary patterns is larger predetermined value.

22. A pattern recognition apparatus according to claim 21 wherein said polygonal approximation means includes:

a polygonal approximation data memory for storing polygonal approximation data on a point pattern required to polygonally-approximate the outline pixels of the digital image;

means for sequentially segmenting the outline pixels of said object region by referring to the polygonal approximation data stored in said polygonal approximation data memory based on the connectivity of said center coordinates to polygon-approximate said outline pixels and determining end points of segments of the polygon;

a polygonal approximation information memory for storing the end point data of said segments as vertex coordinates of the approximation polygon; and

a dictionary pattern memory for storing the content stored in said polygonal approximation information memory when said input image data relates to said dictionary pattern.

23. A pattern recognition apparatus, comprising:

(a) input control means for binarizing input image data by a predetermined threshold;

(b) an image memory for storing said binarized image data;

(c) preprocessing means for identifying an object region from the image data stored in said image memory and concurrently determining position information of a representative point of said region;

(d) a region information memory for storing the position information of said representative point;

(e) outline extraction means for determining a start point of search based on the position information of said representative point, sequentially searching outline pixels of the object region on the image and determining center coordinates of said outline pixels;

(f) an outline memory for storing said center coordinates;

(g) polygonal approximation means for polygonally-approximating the points of said center coordinates within a predetermined allowable error range to polygonally-approximate said object region and thereby provide a polygonal-approximated object region by referring to a table of a pattern series capable of effecting approximation of lines, and sequentially determining vertex coordinates of the polygonally-approximated object region;

(h) a polygonal approximation information memory for storing said vertex coordinates;

(i) region characteristic extraction means for extracting region characteristic data including center information and orientation information about said polygonally-approximated object region from the content of said polygonal approximation information memory;

(j) a region characteristic memory for storing the region characteristic data;

(k) a dictionary pattern memory for storing said vertex coordinates and said region characteristic data when said input image data relates to a dictionary pattern;

(l) common region extraction means for superimposing the polygonal pattern stored in said polygonal approximation information memory on said dictionary pattern stored in said dictionary pattern memory based on the region characteristic data stored in said region characteristic memory and said dictionary pattern memory by transforming the vertex coordinates of said polygonal pattern so that selected axes of both patterns match and determining a common region of both patterns as a sequence of vertex coordinates;

(m) a common region memory for storing the vertex coordinates of said common region; and

(n) recognition means for calculating an area of said common region based on the vertex coordinates stored in said common region memory, determining a ratio of said area to an area of said dictionary pattern or said polygonal pattern and determining the degree of matching of the patterns when said ratio is larger than a predetermined value.

24. A pattern recognition apparatus according to claim 23 wherein said approximation means includes a polygonal approximation data memory for storing polygonal approximation data on the point pattern required to polygonally-approximate the outline pixels of the digital image.

25. A pattern recognition apparatus according to claim 23 wherein said recognition means includes means for calculating a relative position of said polygonal pattern and said dictionary pattern to recognize the center and orientation of said object based on the relative position information.

26. A pattern recognition method, comprising:

(a) a first step for recognizing blobs of the same value, starting points of contours of the blobs, and relations between blobs,

(b) a second step for analyzing inside-outside relations between blobs and an expression of these relations,

(c) a third step for expressing the contours of the blobs by segmented lines according to a look-up table which describes potential mesh patterns to be transformed to a line, (d) a fourth step for calculating x-y coordinates of a center of the object, and a direction of the main axis of secondary moment of the object,

(e) a fifth step for placing a target pattern derived from the object onto a standard pattern by matching the coordinates of the center of said target pattern with coordinates of the center of said standard pattern and transforming vertex coordinates of said segmented lines describing the target pattern on the basis of any angular difference between the direction of said main axis of secondary moment of the object and thereby generating an overlapped polygon, and

(f) a sixth step for identifying a part and its position and orientation by a ratio of the overlapped area to an initial area.