Hybrid template matching for imaging applications

• US 20070031061A1
• Filed: 08/02/2005
• Published: 02/08/2007
• Est. Priority Date: 08/02/2005
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Lee, Chia-Hao, Nagarajan, Ramesh, Tse, Francis
• Granted Patent: US 7,421,147 B2
• Time in Patent Office: Days
• US Class Current: 382/276
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 5/009 Global, i.e. based on properties of the image as a whole applied in cameras using an electronic image sensor H04N23/80, H04N23/70
• CPC Class Codes: G06T 5/92 based on global image properties

Abstract

Training sets consisting of pairs of input images and desired images can be used to produce maps intended for future use in transforming input images into final images. The maps, however, often have improper elements based on incomplete or noisy data. The noisy data and incomplete can be identified via statistics derived from the training set. The map can output a code value instead of a substitution for an improper element. An additional stage of image transformation can identify improper pixels having the code value and replace the code value with an analytic value that is more appropriate.

Claims

1. A method comprising:

obtaining a map comprising at least one map element comprising a map template and a substitution and obtaining an input image comprising at least one input template;

creating a transitional image comprising at least one transitional pixel by transforming each of the at least one input template into a transitional pixel by identifying a match such that the match is one of the at least one map element having a map template matching the input template and copying the match'"'"'s substitution to the transitional pixel; and

creating an output image comprising at least one output pixel by transforming each of the at least one transitional pixel into an output pixel by copying one of at least one analytic value to the output pixel when the transitional pixel matches a code value and copying the transitional pixel to the output pixel otherwise wherein the analytic value is obtained from analysis of the input image, transitional image, or output image.

12. The method of claim 1 wherein a rogue candidate is a candidate having substitution statistics indicating a variance higher than a variance threshold.

13. The method of claim 1 wherein a rogue candidate is a candidate having substitution statistics indicating a spread higher than a spread threshold.

14. The method of claim 1 wherein a rogue candidate is a candidate having substitution statistics indicating symmetry lower than a symmetry threshold.

15. The method of claim 1 wherein a rogue candidate is a candidate having substitution statistics indicating a population less than a population threshold.

16. The method of claim 1 wherein the substitution statistics comprise a mean and further comprising copying the mean to the map substitution.

17. The method of claim 1 wherein the substitution statistics comprise a median and further comprising copying the median to the map substitution.

18. The method of claim 1 wherein the substitution statistics comprise a most common value and further comprising copying the most common value to the map substitution.

19. The method of claim 1 wherein the input image in each of the at least one training pair is a half tone image and the desired image in each of the at least one training pair is a gray scale image.

11. A method comprising:

obtaining a training set comprising at least one training pair comprising a input image and desired image and creating an analysis set comprising at least one candidate comprising a candidate template and substitution statistics comprising at least one statistic; and

classifying each of the at least one candidate as a normal candidate or a rogue candidate; and

creating a map comprising at least one map element comprising a map template and a map substitution wherein each of the at least one candidate is transformed into a map element by a process comprising the steps of copying the candidate template to the map template, copying one of the at least one statistic to the map substitution for each normal candidate, and copying a code value to the map substitution for each rogue candidate.

2. The method of claim 11 wherein the transitional image is created in place from the input image.

3. The method of claim 11 wherein the output image is created in place from the transitional image.

4. The method of claim 11 wherein the input image is a binary image and the output image is a gray scale image.

5. The method of claim 11 further comprising using linear interpolation to find one of the at least one analytic value.

6. The method of claim 11 further comprising using smoothing to find one of the at least one analytic value.

7. The method of claim 11 further comprising using frequency interpolation to find one of the at least one analytic value.

8. The method of claim 11 further comprising using a neighborhood mean as one of the at least one analytic value.

9. The method of claim 11 further comprising using a neighborhood median as one of the at least one analytic value.

10. The method of claim 11 further comprising using a text tag or image tag to guide the choice of the one of at least one analytic value that is copied to the output pixel.

20. A system comprising:

a storage device storing a map comprising at least one map element comprising a map template and a substitution and an input for obtaining an input image comprising at least one input template;

a first module that obtains the map from the storage device, obtains the input image from the input and produces a transitional image comprising at least one transitional pixel by using the map to transform the input image;

a second module that produces a output image from the transitional image by replacing each of the at least one transitional pixel with one of at least one analytic value when the at least one transitional pixel matches a code value.