Dictionary generation for example based image processing

Patent metadata

Field	Value
Publication number	US-10296605-B2
Application number	US-201514968164-A
Country	US
Kind code	B2
Filing date	Dec 14, 2015
Priority date	Dec 14, 2015
Publication date	May 21, 2019
Grant date	May 21, 2019

Abstract

Techniques related to generating dictionaries for example based image processing algorithms are discussed. Such techniques may include iteratively performing example based image processing for candidate look up entries of candidate pairs from a training set database using a current dictionary to determine a test result for each of the look up entries of the candidate pairs and selecting one or more of the candidate pairs for entry in a resultant dictionary based on an error between the test result and a predetermined result entry for the candidate pairs.

First claim

What is claimed is:

1. A method for generating a dictionary for example based image processing comprising:
determining a plurality of candidate pairs from a training set database, wherein each of the candidate pairs comprises a predetermined look up entry and a corresponding predetermined result entry;
performing example based processing for each of the look up entries of the candidate pairs using a current dictionary to determine a test result for each of the look up entries of the candidate pairs;
generating, for each of the look up entries of the candidate pairs, an error result based on the test result and the predetermined result entry;
selecting at least one of the candidate pairs for inclusion in a resultant dictionary based on the error results;
demosaicing a first image based on the resultant dictionary and tracking a usage rate for each entry of the resultant dictionary during the demosaicing of the first image to generate a first usage pattern;
demosaicing a second image based on the resultant dictionary and tracking a second usage rate for each entry of the resultant dictionary during the demosaicing of the second image to generate a second usage pattern; and
determining a similarity scoring of the first image and the second image based on the first and second usage patterns.

2. The method of claim 1 , further comprising:
iteratively determining candidate pairs, performing example based processing, generating error results, and selecting one or more pairs from the candidate pairs for inclusion in the resultant dictionary until the resultant dictionary has reached a maximum number of entries.

3. The method of claim 1 , further comprising:
iteratively determining candidate pairs, performing example based processing, generating error results, and selecting one or more pairs from the candidate pairs for inclusion in the resultant dictionary until no error result in the error results is greater than a threshold for a predetermined number of iterations.

4. The method of claim 1 , further comprising:
iteratively determining candidate pairs, performing example based processing based on a current dictionary, generating error results, and selecting one or more pairs from the candidate pairs for inclusion in the resultant dictionary, wherein the current dictionary at each iteration includes the candidate pairs for inclusion from previous iterations.

5. The method of claim 1 , wherein the candidate pairs comprise pairs of raw image pixel patches and full color pixel patches.

6. The method of claim 1 , wherein the candidate pairs comprise image pixel patches at a first resolution and image pixel patches at a resolution higher than the first resolution.

7. The method of claim 1 , wherein determining the plurality of candidate pairs comprises randomly selecting the plurality of candidate pairs from the training set database.

8. The method of claim 1 , further comprising:
selecting a seed set of pairs from the training set database to provide at least a portion of the current dictionary.

9. The method of claim 1 , wherein selecting the at least one of the candidate pairs for inclusion comprises comparing the error result to a predetermined threshold.

10. A system for approximate nearest neighbor searching comprising:
a memory configured to store a training set database; and
a processor coupled to the memory, the processor to determine a plurality of candidate pairs from a training set database, wherein each of the candidate pairs comprises a predetermined look up entry and a corresponding predetermined result entry, to perform example based processing for each of the look up entries of the candidate pairs using a current dictionary to determine a test result for each of the look up entries of the candidate pairs, to generate, for each of the look up entries of the candidate pairs, an error result based on the test result and the predetermined result entry, to select at least one of the candidate pairs for inclusion in a resultant dictionary based on the error results, to demosaic a first image based on the resultant dictionary and track a usage rate for each entry of the resultant dictionary during the demosaic of the first image to generate a first usage pattern, to demosaic a second image based on the resultant dictionary and track a second usage rate for each entry of the resultant dictionary during the demosaic of the second image to generate a second usage pattern, and to determine a similarity scoring of the first image and the second image based on the first and second usage patterns.

11. The system of claim 10 , wherein the processor is further to iteratively determine candidate pairs, perform example based processing, generate error results, and select one or more pairs from the candidate pairs for inclusion in the resultant dictionary until at least one of the resultant dictionary has reached a maximum number of entries or until no error result in the error results is greater than a threshold for a predetermined number of iterations.

12. The system of claim 10 , wherein the processor is further to iteratively determine candidate pairs, perform example based processing based on a current dictionary, generate error results, and select one or more pairs from the candidate pairs for inclusion in the resultant dictionary, wherein the current dictionary at each iteration includes the candidate pairs for inclusion from previous iterations.

13. The system of claim 10 , wherein the candidate pairs comprise pairs of raw image pixel patches and full color pixel patches.

14. The system of claim 10 , wherein the candidate pairs comprise image pixel patches at a first resolution and image pixel patches at a resolution higher than the first resolution.

15. The system of claim 10 , wherein the processor to select the at least one of the candidate pairs for inclusion comprises the processor to compare the error result to a predetermined threshold.

16. At least one non-transitory machine readable medium comprising a plurality of instructions that, in response to being executed on a device, cause the device to generate a dictionary for example based image processing by:
determining a plurality of candidate pairs from a training set database, wherein each of the candidate pairs comprises a predetermined look up entry and a corresponding predetermined result entry;
performing example based processing for each of the look up entries of the candidate pairs using a current dictionary to determine a test result for each of the look up entries of the candidate pairs;
generating, for each of the look up entries of the candidate pairs, an error result based on the test result and the predetermined result entry; and
selecting at least one of the candidate pairs for inclusion in a resultant dictionary based on the error results;
demosaicing a first image based on the resultant dictionary and tracking a usage rate for each entry of the resultant dictionary during the demosaicing of the first image to generate a first usage pattern;
demosaicing a second image based on the resultant dictionary and tracking a second usage rate for each entry of the resultant dictionary during the demosaicing of the second image to generate a second usage pattern; and
determining a similarity scoring of the first image and the second image based on the first and second usage patterns.

17. The non-transitory machine readable medium of claim 16 further comprising a plurality of instructions that, in response to being executed on a device, cause the device to generate a dictionary for example based image pro

Assignees

• Intel Corp

Inventors

• Barel Hila
• Michael Gilad
• Chalom Edmond

Classifications

• G06V30/1914

  Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries, e.g. user dictionaries · CPC title

• G06V10/763

  Non-hierarchical techniques, e.g. based on statistics of modelling distributions · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06F16/532

  Query formulation, e.g. graphical querying · CPC title

• G06F18/28

  Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries · CPC title