Apparatus and method for use in image processing

• US 5,583,949 A
• Filed: 06/06/1995
• Issued: 12/10/1996
• Est. Priority Date: 03/03/1989
• Status: Expired due to Term

• Current Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Original Assignee: Hewlett-Packard Company (HP Inc.)
• Inventors: Smith, Raymond W., Robson, Christopher J.
• Primary Examiner(s): COUSO, YON JUNG
• Application Number:
  US08/468,517
• Time in Patent Office: 553 Days
• Field of Search: 382/199, 382/198, 382/200, 382/190, 382/203, 382/204, 358/455, 358/456
• US Class Current: 382/199
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/1823 using vector-coding
• CPC Class Codes: G06V 30/196 using sequential comparisons of the image signals with a plurality of references

Abstract

Optical character recognition is achieved by a system which comprises a scanner for scanning a document, an edge extractor for identifying edges in the image produced by the scanner to produce an outline of each object identified in the image, a segmentation facility for grouping the object outlines into blocks, means for identifying features of the outlines, and a final classification stage for providing data in an appropriate format representative of the characters in the image. Also disclosed are a novel edge extractor, a novel page segmentation facility and a novel feature extraction facility.

Claims

1. An optical character recognition system comprising:

scanning means for optically scanning a document to produce a grey level image thereof;

edge extractor means comprising;

identifier means for identifying points along an edge within said grey level image using grey level values so that said points so identified represent substantially the strongest edge;

tracking means for automatically tracking the edge using grey level values to determine if the edge forms a closed loop and if so defining the edge as an outline,

said identifier means identifying alternate points of the edge if the edge does not form a closed loop and said tracking means automatically tracking an alternate edge associated with said alternate points together with at least some of said points on said strongest edge and determining whether the alternate edge forms a closed loop and if so defining the alternate edge as the outline; and

means for producing data indicative of an object based on at least one outline identified in said image, each outline comprising at least a part of one character; and

processing means for processing the data provided by said edge extractor means to produce an output representative of the characters in said image.

2. The system as claimed in claim 1, wherein the outlines identified by said edge extractor means are subjected to polygonal approximation means for processing the data prior to further processing by (1) following the outline of each object to identify a series of points on said outline having a predetermined characteristic, (2) measuring the depth of the concavity between each adjacent pair of said points, and (3) removing concavities having depths less than a predetermined value, whereby the output of said polygonal approximation means approximates the outlines of the characters in said image by a sequence of straight lines.

3. The system as claimed in claim 1, wherein the grey level image comprises a pixel array, said edge extractor being arranged to locate edges by means of an edge operator which generates edge values for each pixel from the grey level value of pixels adjacent to said pixel.

4. The system as claimed in claim 3, wherein the edge operator process a plurality of grey level values and is defined with respect to a 3×

3 matrix neighborhood of x as follows;

space="preserve" listing-type="tabular">______________________________________ 5 6 7 4 X 0 3 2 1 ______________________________________

in which the numbers represent the chain codes of adjacent points, the edge value of an adjacent point with chain code n being;

grey level (n+1) (modulo

     8)-grey level (n-1) (modulo

     8).

5. The system as claimed in claim 4, wherein each of the edges is traced in accordance with an algorithm in which, for a point with chain code n, edge values are derived for points with chain codes n-2, n-1, n, n+1, and n+2 from the current position X, the highest edge value is marked as the current position X and defines the strongest point, and the process is repeated to identify and mark successive strongest points.

6. The system as claimed in claim 1, wherein said processing means includes a page segmentation facility for arranging the data indicative of the outlines into a desired order and means for identifying particular features of said outlines.

7. The system as claimed in claim 6, wherein said page segmentation facility identifies groups of said outlines to form said outlines into blocks of outlines and arranges said blocks into said desired order.

8. The system as claimed in claim 7, wherein the page segmentation facility can identify blocks as text or non-text blocks by determining whether the proportion of spurious characters in the block exceeds a defined threshold value.

9. The system as claimed in claim 8, wherein the block type is determined on the basis of a classification of each outline.

10. The system as claimed in claim 9, wherein data representative of the classification of each outline is provided by an object outline classification process.

11. The system as claimed in claim 8, wherein the block order is determined in accordance with the manner in which the blocks nest.

12. The system as claimed in claim 6, wherein the output of said page segmentation facility comprises data representative of at least one possible class to which the object represented by said at least one outline belongs together with a weighting value, said weighting value indicating the probability that the outline is associated with a particular class.

13. The system as claimed in claim 1, wherein said edge extractor means further comprises a feature extraction facility for processing said data to identify particular features of each outline and to classify the outlines into classes in accordance with the identified features.

14. The system as claimed in claim 13, wherein the features comprise any one of concavity, closure, axis, symmetry or line either singly or in combination.

15. The system as claimed in claim 6, wherein said processing means includes a final classification stage for making the final classification of each of said outlines, wherein said final classification stage operates on the data output by the page segmentation facility using lexical rules, Markov chain analysis and dictionary search techniques.

16. The system as claimed in claim 15, wherein the output of said final classification stage is in ASCII code.

17. The system as claimed in claim 1, wherein the operation of the edge extractor derives an edge value for a pixel based on the grey level values of pixels immediately adjacent to the pixel based on the direction of tracking.

18. An image processing system comprising an edge extractor for automatically producing data indicative of the outline of objects in a grey level image generated by an optical scanning device, wherein said outline comprises at least part of one or more characters, said grey level image comprising a pixel array, said edge extractor comprising means for sequentially scanning said pixel array to locate a first strongest edge and for tracing the outline of one of the objects starting with said first strongest edge and by locating a plurality of additional strongest edges contiguous with one another to determine whether or not the entire outline can be traced using only said first and additional strongest edges and automatically locating alternate edges that are used to complete the entire outline when said additional strongest edges do not complete the outline, said edge extractor further comprising edge operator means for generating edge values for each pixel using grey level values, wherein the operation of the edge extractor is such that the edge value for a current pixel is derived from the relative values of the grey level values of the pixels which are immediately adjacent to the current pixel based on the direction of tracing.

19. The system as claimed in claim 18, wherein the edge operator processes a plurality of grey level values and is defined with respect to a 3×

3 neighborhood matrix of x as follows;

##STR1## in which the members represent the chain codes of adjacent points, the edge value of an adjacent point with chain code n being;

grey level (n+1) (modulo

     8)-grey level (n-1) (modulo

     8).

20. The system as claimed in claim 19, wherein said outline is traced in accordance with an algorithm in which for a point with chain code n, edge values are derived for points with chain codes (n-2), (n-1), n, (n+1), and (n+2) from the current position, the point with the strongest edge values is marked, and the process is repeated to identify and mark successive strongest edge values.

21. The system according to claim 18, wherein the edge extractor further comprises a polygonal approximation facility for approximating the outline of an object generated by the edge extractor by at least one polygon.

22. An edge extractor according to claim 18, wherein the edge outline is represented by a closed loop.

23. An image processing system comprising means for scanning a document to generate an image represented by grey level values and automatically producing data representative of an outline of an object in the image comprising an edge extractor for tracing said outline by generating an edge value for a pixel derived from the grey level values of the pixels which are immediately adjacent to said pixel based on the direction of tracing, said edge extractor locating a strongest edge based on said edge values and tracing said outline by locating additional strongest edges if said additional strongest edges in combination complete the outline, said edge extractor automatically locating an alternate edge to complete the outline when said additional strongest edges do not complete the outline, said system further comprising a page segmentation facility for identifying groups of outlines to form said groups into blocks of outlines and for arranging said blocks into order, wherein said outlines are representative of at least a portion of one or more characters.

24. The system as claimed in claim 23, wherein the page segmentation facility can identify blocks as text or non-text by determining whether a proportion of spurious characters in the block exceeds a defined threshold value.

25. The system as claimed in claim 23, wherein the block type is determined on the basis of a classification of each outline.

26. The system as claimed in claim 25, wherein the data representing the classification of each outline is provided by an object outline classification process.

27. The system as claimed in any claim 24, wherein the block order is determined in accordance with the manner in which the blocks nest.

28. An image processing system comprising means for scanning a document to generate an image represented by grey level values and automatically producing output data representative of an outline of an object in the image, the system comprising an edge extractor for tracing said outline by generating an edge value for a pixel derived from the grey level values of pixels immediately adjacent to said pixel relative to the direction of tracing, said edge extractor locating a strongest edge based on said edge values and tracing said outline by locating additional strongest edges if said additional strongest edges in combination complete the outline, said edge extractor automatically locating an alternate edge to complete the outline when said additional strongest edges do not complete the outline, said system further comprising a facility for processing said data to identify particular features of each outline and to classify the outlines into classes of characters likely to represent the character for each outline in accordance with the identified features, and for classifying as spurious outlines those outlines not representing a character, wherein the features identified comprise one or more of concavity, closure, axis, symmetry and line.

29. The system as claimed in claim 28, wherein said output data is produced by said edge extractor.

30. The system as claimed in claim 28, wherein the classification data includes a rating or weighting value indicative of the strength of the match of the outline to the class with which the object has been associated.

31. A method of processing a document to produce data representative of characters on said document, said method comprising the steps of:

scanning said document to produce a pixel array of grey level values representing a grey level image of the document;

scanning the grey level values and for each grey level value comparing the grey level values of pixels immediately on either side of the pixel being scanned;

locating a first edge associated with one of the pixels based on the grey level values so compared;

automatically locating additional edges and tracing a strongest edge in said grey level image representing an outline comprising at least a portion of one or more characters in said image by comparing the grey level values immediately adjacent the pixel associated with each said additional edge; and

determining whether or not said strongest edge forms a complete outline;

automatically identifying an alternate edge that does form a complete outline when said strongest edge does not form a complete outline, wherein the alternate edge is identified by comparing grey level values;

processing data representing said strongest and said alternate edges that form complete outlines to provide data representative of the characters in said image, said processing step comprising automatically segmenting and classifying the data into blocks representative of a plurality of outlines, each outline comprising at least a portion of one or more characters.

32. The method as claimed in claim 31, wherein the data is subjected to a polygonal approximation step.

33. The method as claimed in claim 31, wherein said processing step comprises identifying a feature or features of said complete outlines and providing data representative of a class to which the object represented by said complete outlines belong.

34. A method as claimed in claim 33, wherein said features include concavities and closures.