Clustering

• US 7,164,797 B2
• Filed: 04/25/2002
• Issued: 01/16/2007
• Est. Priority Date: 04/25/2002
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Malvar, Henrique S., Simard, Patrice Y., Renshaw, Erin L.
• Primary Examiner(s): Mancuso; Joseph
• Assistant Examiner(s): SCHAFFER, JONATHAN C
• Application Number:
  US10/133,558
• Publication Number:
  US 20030202709A1
• Time in Patent Office: 1,727 Days
• Field of Search: 382/243, 382/218, 382/225
• US Class Current: 382/218
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/18 Extraction of features or characteristics of the image
• CPC Class Codes: G06V 30/19013 Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching specially adapted for image segmentation G06T7/10; specially adapted for the analysis of motion G06T7/20; specially adapted for image alignment G06T7/30; specially adapted for the calculation of depth from stereo images G06T7/50; specially adapted for position determination G06T7/70
• CPC Class Codes: G06V 30/40 Document-oriented image-based pattern recognition

Abstract

Systems and methods for performing clustering of a document image are disclosed. A property of an extracted mark from a document is compared to the properties of the existing clusters. If the property of the mark fails to match any of the properties of the existing clusters, the mark is added as a new cluster to the existing cluster. One property that can be utilized is x size and y size, which is the width and height, of the existing clusters. Another property that can be employed is ink size, which refers to the ratio of black pixels to total pixels in a cluster. Yet another property that can be utilized is a reduced mark or image, which is a pixel size reduced version the bitmap of the mark and/or cluster. The above properties can be employed to identify mismatches and reduce the number of bit by bit comparisons performed.

Claims

1. A clustering system comprising:

a mark extractor to extract a mark from a document;

a match component operative to compare one or more properties of the mark to one or more match properties of existing clusters of marks so as to identify matching existing clusters, the match properties comprising resized images for the existing clusters, and the match component operative to compute a range of acceptable values for the one or more properties using a threshold;

a two dimensional table that stores the existing clusters according to box size, wherein if no matches are identified, the mark is added to the existing clusters as a new cluster, and if a match is identified then bitmaps of the mark and the matching existing clusters are compared; and

a match symbol component operative to compare a bitmap of the mark to a bitmap of the matching existing clusters the bitmaps are compared bit by bit to identify a matching cluster having a similar bitmap, wherein once an acceptable matching cluster is identified based on the bitmaps, the bitmap of the matched cluster is updated with an average based on the bitmap of the mark, and if no matching cluster is acceptable based on the bitmaps, the mark is added to the existing clusters as a new cluster.

2. The clustering system of claim 1, further comprising a library to store the existing clusters.

3. The system of claim 1, further comprising a table of match properties of the existing clusters.

4. The clustering system of claim 2, the library comprising a local library and a global library, the local library comprising existing clusters added from a current page of the document and the global library comprising existing clusters added from previous pages of the document.

5. The clustering system of claim 1, the mark extractor being operative to locate the mark and extract the mark from the document.

6. The clustering system of claim 1, the match properties further comprising x sizes and y sizes for the existing clusters.

7. The clustering system of claim 1, the match properties further comprising ink sizes for the existing clusters.

8. The clustering system of claim 6, the at least one property of the mark being x size and y size.

9. The clustering system of claim 7, the at least one property of the mark being ink size.

10. The clustering system of claim 1, the at least one property of the mark being a resized mark image.

11. The clustering system of claim 1, the match component further operative to add the mark as a new cluster on the mark mismatching the existing clusters.

12. The clustering system of claim 1, the match symbol component being further operative to add the mark to a set of matching marks of the most matching cluster.

13. The clustering system of claim 11, further comprising a library, the library being operative to store the existing clusters and to sort the existing clusters according to a frequency of occurrence for the existing clusters.

14. A photocopier employing the clustering system of claim 1.

15. A fax machine employing the clustering system of claim 1.

16. A digital camera employing the clustering system of claim 1.

17. An image encoding system employing the clustering system of claim 1.

18. A method of clustering comprising:

locating a mark within a document;

comparing a first property of the mark with first properties of existing clusters to identify matching and mismatching clusters by computing an acceptable range of properties;

on a match of the first property, comparing a bitmap of the mark with bitmaps of the matching clusters, the bitmaps are compared bit by bit to identify a matching cluster having a similar bitmap, wherein once an acceptable matching cluster is identified based on the bitmaps, the bitmap of the matched cluster is updated with an average based on the bitmap of the mark, and if no matching cluster is acceptable based on the bitmaps, the mark is added to the existing clusters as a new cluster;

on a mismatch of the first property, adding the mark as the new cluster to the existing clusters; and

generating a resized mark image from the bitmap of the mark.

19. The method of claim 18, the first property being compared comprising at least one of the group comprising x size, y size, ink size and resized mark image.

20. The method of claim 18, the resized mark image generated by dividing the bitmap of the mark into 9 regions and using a majority of pixels in each of the 9 regions for corresponding pixels of a 9 pixel, 3 by 3, resized mark image.

21. The method of claim 18, further comprising computing a hot point for the mark.

22. The method of claim 18, further comprising comparing a second property of the mark with second properties of existing clusters to identify the matching and the mismatching clusters.

23. The method of claim 22, further comprising comparing a third property of the mark with third properties of existing clusters to identify the matching and the mismatching clusters.

24. The method of claim 18, wherein comparing a bitmap of the mark with bitmaps of the matching clusters comprises computing a center of gravity of the mark and a center of gravity of the cluster, computing an xor distance between the mark the matching clusters and selecting the most matched cluster having a smallest xor distance with the mark and the smallest xor distance being within an acceptable range.

25. The method of claim 18, wherein frequently occurring existing clusters are maintained in a global library and rarely occurring clusters are maintained in a local library.

26. A document encoding system comprising:

a mask separator operative to generate a binary mask from a document image, the binary mask including textual information;

a background foreground segmenter operative to segment a foreground image and a background image from the document image according to the binary mask; and

a clustering system operative to identify clusters in the mask in a computationally efficient manner.

27. A computer readable medium storing computer executable components operable to perform a method of clustering, comprising:

a component for locating a mark;

a component for comparing a first property of the mark with first properties of existing clusters to identify matching and mismatching clusters by computing an acceptable range of properties;

on a match of the first property, a component for comparing a bitmap of the mark with bitmaps of the matching clusters, the bitmaps are compared bit by bit to identify a matching cluster having a similar bitmap, wherein once an acceptable matching cluster is identified based on the bitmaps, the bitmap of the matched cluster is updated with an average based on the bitmap of the mark, and if no matching cluster is acceptable based on the bitmaps, the mark is added to the existing clusters as a new cluster;

on a mismatch of the first property, a component to add the mark as a new cluster to the existing clusters; and

a component for generating a resized mark image from the bitmap of the mark.

28. A computer readable medium storing computer executable instructions operable to perform a method of clustering, comprising:

for each page of at least one page of a document;

a component for finding at least one mark;

a component for comparing a first property of the at least one mark with first properties of existing clusters to identify matching and mismatching clusters by computing a range of acceptable values for the first property using a threshold;

on a match of the first property, a component for comparing a bitmap of the at least one mark with bitmaps of the matching clusters, the bitmaps are compared bit by bit to identify a matching cluster having a similar bitmap, wherein once an acceptable matching cluster is identified based on the bitmaps, the bitmap of the matched cluster is updated with an average based on the bitmap of the mark, and if no matching cluster is acceptable based on the bitmaps, the mark is added to the existing clusters as a new cluster; and

on a mismatch of the first property, a component for adding the at least one mark as a new cluster to the existing clusters;

a component for updating a global library; and

a component for generating a resized mark image from the bitmap of the mark.