Finding text regions from coloured image independent of colours

• US 20120155754A1
• Filed: 12/13/2011
• Published: 06/21/2012
• Est. Priority Date: 12/17/2010
• Status: Active Grant

• Current Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Original Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Inventors: Chen, Yu-Ling, Ballu, Kailash Nandish
• Granted Patent: US 8,503,781 B2
• Time in Patent Office: Days
• US Class Current: 382/164
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/18076 by analysing connectivity, e.g. edge linking, connected component analysis or slices
• CPC Class Codes: G06V 30/413 Classification of content, e.g. text, photographs or tables

Abstract

Methods (1200), apparatuses, and computer program for finding a region containing text in a colour bitmap image comprising pixels are provided. Connected components (CCs) are generated (1200) from the colour bitmap image by grouping substantially similarly coloured and adjacent pixels. Independently of colour, which of the connected components are text connected components are determined (1212) dependent upon attributes of the generated connected components. For each text CC) a value is assigned (1214) to each tile that comprises at least a portion of the text connected component. The value indicates the presence of text overlapping the tile. Each tile comprises pixels of the colour bitmap image. The region containing text in the colour bitmap image is determined (1216) dependent upon the assigned values of the tiles.

Claims

1. A method of finding a region containing text in a colour bitmap image comprising a plurality of pixels, the method comprising the steps of:

generating connected components (CCs) from the colour bitmap image by grouping substantially similarly coloured and adjacent pixels;

determining, independently of colour, which of the connected components are text connected components dependent upon attributes of the generated connected components;

assigning, for each text connected component (CC), a value to each tile that comprises at least a portion of the text connected component, said value indicating the presence of text overlapping the tile, each tile comprising a plurality of pixels of said colour bitmap image; and

determining the region containing text in the colour bitmap image dependent upon said assigned values of the tiles.

2. The method as claimed in claim 1, comprising dividing the colour bitmap image into a plurality of tiles.

3. The method as claimed in claim 1, wherein said value indicating the presence of text overlapping the tile comprises a text pixel density of the text connected component in the tile, or a text CC count density of the text connected component in the tile.

4. The method as claimed in claim 1, wherein said assigning step comprises:

initialising geometric distribution statistics contained in each tile;

for each text CC, generating and distributing statistical measures pertaining to the current text CC into the geometric distribution statistics of each tile that is overlapped by the current text CC; and

for each tile, finalising the geometric distribution statistics of the current tile.

5. The method as claimed in claim 4, wherein said generating and distributing step comprises:

calculating the number of overlapped tiles;

dividing a text pixel count by said number of overlapped tiles;

accumulating said divided text pixel count to a corresponding text total pixel count in all overlapped tiles;

accumulating a fraction of one divided by said number of overlapped tiles to the corresponding text CC count in all overlapped tiles; and

updating a corresponding minimum pixel count of a text CC and a maximum pixel count of a text CC by said divided text pixel count in all overlapped tiles.

6. The method as claimed in claim 5, wherein said finalising step comprises:

setting the average number of pixels per text CC in the tile by the value of dividing the total text pixel count by text CC count in each tile; and

setting the tile text pixel density by the value of dividing the total text pixel count by the total number of pixels in the tile.

7. The method as claimed in claim 1, wherein the region determining step comprises dividing an initial region into one or more regions comprising text dependent upon average pixel densities along rows and columns of tiles in a text region.

8. The method as claimed in claim 7, wherein region determining step comprises:

creating an initial text region; and

wherein said dividing step comprises, for each region;

calculating average pixel densities along rows and columns of tiles in the current text region;

determining if the current text region should be split dependent upon the calculated average pixel densities; and

splitting the current region into at least two regions if the current region is determined should be split.

9. The method as claimed in claim 1, wherein the region determining step comprises growing one or more text regions comprising text using one or more tiles comprising text.

10. The method as claimed in claim 9, wherein region determining step comprises:

creating an empty list to hold text regions; and

wherein said growing step comprises, for each tile containing text;

generating a list of probable text regions that the current tile might belong to;

providing a text region dependent upon the list of probable text regions; and

adding the current tile to the text region and adding the text region to the list to hold text regions.

11. The method as claimed in claim 1, further comprising the step of refining text connected components dependent upon noise characteristics, wherein text and noise in said colour bitmap image are discriminated.

12. The method as claimed in claim 11, wherein said noise comprises at least one of watermark, texture, and halftone noise.

13. The method as claimed in claim 11, wherein said text CC refining step comprises

receiving geometric distribution statistics for each tile;

and

determining noise characteristics of said colour bitmap image dependent upon geometric distribution statistics in each tile, the noise in said colour bitmap image being either primarily watermark noise or non-watermark noise.

14. The method as claimed in claim 13, wherein said noise characteristics determining step comprises:

profiling said geometric distribution statistics in each tile to determine characteristics of noise; and

identifying locations of noise based on said characteristics of noise.

15. The method as claimed in claim 14, further comprising reclassifying text CCs matching said noise characteristics at said locations.

16. The method as claimed in claim 14, wherein said profiling step comprises:

forming a histogram on an average text pixel count per text CC in each tile, wherein bins are partitioned to capture characteristics of noise; and

collecting statistics in each bin including at least one of maximum text pixel count per text CC of an included tile, included tile count, mean of text pixel count per text CC per included tile, standard deviation of text pixel count per text CC per included tile, mean of text CC count per included tile and standard deviation of text CC count per included tile.

17. The method as claimed in claim 14, wherein said identifying step comprises:

determining the type of majority noise based on an identity of the largest bin, the type of noise being categorised into watermark noise and non-watermark noise;

deriving at least one text CC count threshold and at least one text pixel count per text CC threshold from said statistics of the largest bin if the type of majority noise is determined to be watermark noise;

marking tiles with said geometric statistics satisfying said at least one text CC count threshold and at least one text pixel count per text CC threshold as noisy tiles;

marking tiles adjacent to said noisy tiles with said geometric statistics close to said at least one text CC count threshold and at least one text pixel count per text CC threshold as noisy tiles;

marking tiles with said geometric statistics indicating a dense cluster of small text CCs or a sparse cluster of very small text CCs as noisy tiles if the type of majority noise is determined to be non-watermark noise.

18. The method as claimed in claim 15, wherein said matching step checks each text CC for combinations of these characteristics:

a pixel count lower than said at least one pixel count per text CC threshold if the type of majority noise is determined to be watermark noise;

a pixel count lower than a predetermined pixel count threshold if the type of majority noise is determined to be non-watermark noise; and

a bounding box overlapping any noisy tiles for both noise types or a bounding box surrounded by more than three (3) noisy tiles if the type of majority noise is determined to be watermark noise.

19. An apparatus for finding a region containing text in a colour bitmap image comprising a plurality of pixels, the apparatus comprising:

a memory for storing data and a computer program; and

a processor unit coupled to the memory for executing a computer program, said memory and said processor configured to find the region containing text in the colour bitmap image, the computer program comprising;

computer program code means for generating connected components (CCs) from the colour bitmap image by grouping substantially similarly coloured and adjacent pixels;

computer program code means for determining, independently of colour, which of the connected components are text connected components dependent upon attributes of the generated connected components;

computer program code means for assigning, for each text connected component (CC), a value to each tile that comprises at least a portion of the text connected component, said value indicating the presence of text overlapping the tile, each tile comprising a plurality of pixels of said colour bitmap image; and

computer program code means for determining the region containing text in the colour bitmap image dependent upon said assigned values of the tiles.

20. A computer readable storage medium having recorded therein a computer program for finding a region containing text in a colour bitmap image comprising a plurality of pixels, the computer program comprising:

computer program code means for generating connected components (CCs) from the colour bitmap image by grouping substantially similarly coloured and adjacent pixels;

computer program code means for determining, independently of colour, which of the connected components are text connected components dependent upon attributes of the generated connected components;

computer program code means for assigning, for each text connected component (CC), a value to each tile that comprises at least a portion of the text connected component, said value indicating the presence of text overlapping the tile, each tile comprising a plurality of pixels of said colour bitmap image; and

computer program code means for determining the region containing text in the colour bitmap image dependent upon said assigned values of the tiles.