Skew Correction for Scanned Japanese/English Document Images

• US 20110052094A1
• Filed: 08/28/2009
• Published: 03/03/2011
• Est. Priority Date: 08/28/2009
• Status: Abandoned Application

• Current Assignee: Seiko Epson Corporation (Seiko Group)
• Original Assignee: Seiko Epson Corporation (Seiko Group)
• Inventors: Xiao, Jing, Gao, Chunyu
• Application Number:
  US12/549,854
• Publication Number:
  US 20110052094A1
• Time in Patent Office: Days
• US Class Current: 382/296
• CPC Class Codes: G06V 30/1478 of characters or characters lines

Abstract

Bit data is encoded into characters in an input image by providing different darkness levels for pixels in either half of a text character. In order to detect document skew angle error prior to attempting to extract an encoded message, the orientation angle of text line(s) in the input image is determined, and a document skew angle is estimated from the line orientation angle(s). A Hough transform may be used to estimate line orientation angle. To determine if an input image of a document is orientated horizontally or vertically, the input image may be scanned for text characters along a first reference angle, and along a second angle 90° to the first angle. The scan that provides the higher number of potential text character is selected as the correct orientation for the input image.

Claims

1. A method for applying skew correction to an input image including providing a processing device to implement steps comprising:

identifying an alignment of consecutive characters on said input image defining a character line;

estimating a line orientation angle of said character line;

defining a document skew angle from said line orientation angle;

rotating said input image by an amount determined to counteract said document skew angle.

2. The method of claim 1, wherein a plurality of said character lines are identified in said input image.

3. The method of claim 2, wherein said line orientation angle is estimated for only the longest one of said plurality of character lines.

4. The method of claim 2, wherein each of said plurality of character lines has a respectively estimated line orientation angle.

5. The method of claim 4, wherein said document skew angle is defined as the mean of said respectively estimate line orientation angles.

6. The method of claim 1, wherein said step of identifying an alignment of consecutive characters includes the following:

(a) a site detection sub-step for identifying a bounding box around each of said characters;

(b) a centroid estimation sub-step for determining a centroid for each bounding box; and

(c) a Hough transform sub-step for applying a Hough transform to the centroids to identify a lines that traverse said centroids.

7. The method of claim 6, wherein said step of estimating a line orientation angle includes:

a text line orientation step for extracting, from said Hough transform, a line orientation angle from at least one of said lines that traverse said centroids.

8. The method of claim 6, wherein in sub-step (b), for each bounding box:

a first side of the bounding box extends from a first location X1 to a second location X2 along a first axis of a Cartesian coordinate system;

a second side of the bounding box that intercepts the first side extends from a third location Y1 to a fourth Y2 along a second axis of said Cartesian coordinate system; and

said centroid estimation step includes determining Cartesian coordinates (Cx, Cy) of the centroid of the boundary box, as Cx=(X2+X1)/2 and Cy=(Y2+Y1)/2.

9. The method of claim 8, wherein said Hough transform sub-step (c) includes:

for each centroid, determining a radial vector r parameter and an angle θ

parameter by consecutively sweeping radial vector r from an origin across the input image and determining the angle θ

where radial vector r intercepts each centroid.

10. The method of claim 8, wherein the radial vector r of each centroid is defined as r=(Cx) cos θ

+(Cy)sin θ

.

11. The method of claim 9, wherein radial vector r is shifted at discrete angle intervals during each sweep, and the length of radial vector r is incremented after each sweep.

12. The method of claim 11, wherein increments in radial vector r are selected in the range of 1-20 pixels in proportion to the font size of said characters.

13. The method of claim 11, wherein said discrete angle intervals are selected between 0.01 degrees to 0.5 degrees.

14. The method of claim 11, wherein radial vector r is swept from between −

90 degrees and 90 degrees.

15. The method of claim 6, wherein said Hough transform sub-step (c) includes:

plotting radial vector r versus angle θ

on a Hough Map for all character sites;

identifying as peaks any interception points of groups of lines on said Hough Map, each peak corresponding to a character line on said input image;

identifying the angle θ

coordinate of each peak as the line orientation angle of its corresponding character line.

16. The method of claim 15, wherein the document skew angle is defined as angle θ

coordinate of the peak with the maximum value in the Hough Map.

17. The method of claim 15, wherein the document skew angle is defined as the line orientation angle of the longest character line in said input image.

18. The method of claim 1, wherein said characters are text characters.

19. The method of claim 1, wherein said text characters are Eastern writing characters.

20. An image processing device for applying skew correction to an input image, comprising:

an input for receiving said input image;

a site detection unit for identifying bounding boxes around characters in said input image;

a centroid estimation unit for identifying a centroid of each bounding box;

a text line orientation unit for determining an orientation angle for at least one line of characters within said input image;

a document orientation unit for determining a document skew angle for said input image based on the angle orientation results of said text line orientation unit;

an orientation correction unit for rotating said input image by an amount determined to counteract said document skew angle.

21. The image processing device of claim 21, wherein said at least one line of characters is the longest line of characters within said input image, and said document skew angle is set equal to the orientation angle of the longest line of characters in said input image.

22. The image processing device of claim 21, wherein text line orientation unit determines the orientation angles of a plurality of lines of characters within said input image, and said document orientation unit determines the document skew angle as the mean of said orientation angles.

23. The image processing device of claim 21, wherein said a text line orientation unit:

determines a radial vector r parameter and an angle 0 for each centroid and applies a Hough transform to the centroids;

plots radial vector r versus angle θ

on a Hough Map for all centroids;

identifies as peaks any interception points of groups of lines on said Hough Map, each peak corresponding to a character line on said input image;

identifies the angle θ

coordinate of each peak as the line orientation angle of its corresponding character line.

24. The image processing device of claim 23, wherein said document orientation unit defines the document skew angle as the angle θ

coordinate of the peak with the maximum value in the Hough Map.