Control of computer vision pre-processing based on image matching using structural similarity

Patent metadata

Field	Value
Publication number	US-9613294-B2
Application number	US-201514663242-A
Country	US
Kind code	B2
Filing date	Mar 19, 2015
Priority date	Mar 19, 2015
Publication date	Apr 4, 2017
Grant date	Apr 4, 2017

Abstract

A processor computes a measure of input image structural complexity of an input image, and searches a database of true positives to find one or more entries in the database that represent true positive images that are structurally similar to the input image. The processor compares a measure of signal quality of the input image and a measure of signal quality of one of the true positive images, as retrieved from the database, and based on the comparison updates a control variable that configures a signal quality conditioning process that is to be performed on the input image prior to processing of the input image by a computer vision processor thus improving performance of the computer vision task. Other embodiments are also described and claimed.

First claim

What is claimed is:

1. An image processing method for a computer vision system, comprising:
processing an input image to compute a measure of input image structural complexity, of the input image;
using a) the measure of input image structural complexity and b) a measure of true positive structural complexity, of a true positive image, computing a measure of structural similarity of the input image to the true positive image, wherein the true positive image is one that has resulted in a correct decision by a computer vision processor; and
in response to the measure of structural similarity indicating sufficient similarity, performing a comparison between a measure of signal quality of the input image and a measure of signal quality of the true positive image and, based on the comparison, updating a control variable that configures a signal quality conditioning process that is performed on the input image prior to processing of the input image by a computer vision processor.

2. The method of claim 1 wherein the measure of input image structural complexity comprises an input image color or graylevel complexity metric, an input image lightness complexity metric, and an input image spatial complexity metric.

3. The method of claim 2 wherein the measure of true positive structural complexity comprises a true positive color or graylevel complexity metric, a true positive lightness complexity metric, and a true positive spatial complexity metric.

4. The method of claim 3 wherein the measure of structural similarity is a function of two or more of i) the input image and true positive color or graylevel complexity metrics, ii) the input image and true positive lightness complexity metrics and iii) the input image and true positive spatial complexity metrics.

5. The method of claim 1 further comprising accessing a database having a plurality of entries, wherein each entry comprises a plurality of true positive structural complexity metrics associated with a plurality of true positive signal quality metrics,
wherein the plurality of true positive structural complexity metrics were derived from a respective one or more true positive images, and the plurality of true positive signal quality metrics were derived from the respective one or more true positive images,
wherein the database is accessed to retrieve one or more of the plurality of true positive structural complexity metrics in a given entry,
and wherein computing the measure of structural similarity uses a) one or more of a plurality of input image complexity metrics derived from the input image, and b) said one or more of the plurality of true positive structural complexity metrics retrieved from the given entry in the database.

6. The method of claim 5 wherein the plurality of input image complexity metrics comprise an input image color or graylevel complexity metric, an input image lightness complexity metric and an input image spatial complexity metric,
and wherein computing the measure of structural similarity comprises computing a structural similarity metric that is a function of
a) two or more of i) the input image color or graylevel complexity metric, ii) the input image lightness complexity metric and iii) the input image spatial complexity metric, and
b) a corresponding two or more of the plurality of true positive structural complexity metrics retrieved from the given entry in the database.

7. The method of claim 1 wherein updating the control variable, that configures a signal quality conditioning process that is performed on the input image prior to processing of the input image by a computer vision processor, comprises increasing or decreasing the control variable in proportion to how much a) the measure of signal quality of the true positive image, is greater than or less than b) the measure of signal quality of the input image.

8. An image processing system, comprising:
a processor; and
memory having stored therein instructions that when executed by the processor
compute a measure of input image structural complexity, of an input image;

compute a measure of signal quality of the input image;
retrieve one or more of the plurality of true positive structural complexity metrics in a given entry of the database, and compute a structural similarity metric using a) the measure of input image structural complexity and b) the one or more true positive structural complexity metrics retrieved from the database; and
perform a comparison between the measure of signal quality of the input image and the true positive signal quality metric retrieved from the database, wherein in response to the structural similarity metric indicating sufficient similarity, a control variable is updated in accordance with the comparison, wherein the control variable configures a pre-processor of a computer vision system.

9. The system of claim 8 wherein the measure of input image structural complexity comprises an input image color or graylevel complexity metric, an input image lightness complexity metric, and an input image spatial complexity metric.

10. The system of claim 9 wherein the plurality of true positive structural complexity metrics comprise a true positive color or graylevel complexity metric, a true positive lightness complexity metric, and a true positive spatial complexity metric.

11. The system of claim 10 wherein the measure of structural similarity is a function of two or more of i) the input image and true positive color or graylevel complexity metrics, ii) the input image and true positive lightness complexity metrics and iii) the input image and true positive spatial complexity metrics.

12. The system of claim 11 wherein the pre-processor comprises two or more of a brightness processor, a sharpness processor, contrast processor, and a noise reduction processor, that are configurable by two or more control variables, respectively.

13. The system of claim 11 wherein the memory has stored therein instructions that when executed by the processor compute the structural similarity metric using a) one or more, but not all, of a plurality of input image complexity metrics that have been derived from the input image, and b) a corresponding one or more, but not all, of the true positive structural complexity metrics in the given entry in the database.

14. The system of claim 11 wherein the memory has stored therein instructions that when executed by the processor update the control variable in accordance with the comparison, by increasing or decreasing the control variable in proportion to how much a) the measure of signal quality of the true positive image, is greater than or less than b) the measure of signal quality of the input image.

15. The system of claim 11 wherein the processor and memory are in a server, and wherein the server is to receive the input image from an end user platform or from a file storage via communication over the Internet.

16. The system of claim 11 in combination with a server, wherein the server is to provide the received input image to the computer vision processor and to the processor, and then serve performance data from the computer vision processor, wherein the performance data is generated by the computer vision processor based on having operated upon the input image, over the Internet.

17. An article of manufacture for a computer vision system, comprising
a non-transitory computer readable medium having stored thereon instructions that program a processor to compute a measure of input image structural complexity of an input image, and search a database of true positives to find

Assignees

• Intel Corp

Inventors

• Caviedes Jorge E
• Wu Sin Lin

Classifications

• G06F18/217 Primary

  Validation; Performance evaluation; Active pattern learning techniques · CPC title

• G06F18/22

  Matching criteria, e.g. proximity measures · CPC title

• G06K9/2036

  Physics · mapped topic

• G06K9/00624

  Physics · mapped topic

• G06K9/6201 Primary

  Physics · mapped topic