Methods and systems for increasing facial recognition working rang through adaptive super-resolution

Patent metadata

Field	Value
Publication number	US-9384386-B2
Application number	US-201414472981-A
Country	US
Kind code	B2
Filing date	Aug 29, 2014
Priority date	Aug 29, 2014
Publication date	Jul 5, 2016
Grant date	Jul 5, 2016

Abstract

Disclosed herein are methods and systems for increasing facial-recognition working range through adaptive super-resolution. One embodiment takes the form of a process that includes calculating one or more video metrics with respect to an input set of video frames. The process also includes obtaining a metric-specific weighting factor for each of the calculated video metrics. The process also includes calculating a weighted sum based on the obtained metric-specific weighting factors and the corresponding calculated video metrics. The process also includes selecting, based at least in part on the calculated weighted sum, a super-resolution technique from among a plurality of super-resolution techniques. The process also includes outputting an indication of the selected super-resolution technique.

First claim

What is claimed is:

1. A method for selecting a super-resolution technique for improved face recognition performance, the method including:
calculating one or more video metrics with respect to an input set of video frames;
obtaining a metric-specific weighting factor for each of the calculated video metrics;
calculating a weighted sum based on the obtained metric-specific weighting factors and the corresponding calculated video metrics;
selecting, based on the calculated weighted sum, a particular super-resolution technique from among a plurality of pre-determined super-resolution techniques by referencing correlation data that maps ranges of weighted sums to corresponding super-resolution techniques; and
using the selected super-resolution technique to generate a super-resolved set of video frames by applying the selected super-resolution technique to the input set of video frames.

2. The method of claim 1 , further comprising selecting one or more configuration parameters for the selected particular super-resolution technique based on one or more of the calculated video metrics.

3. The method of claim 1 , wherein the one or more calculated video metrics includes a temporal-motion metric.

4. The method of claim 3 , further comprising selecting a super-resolution regularization configuration parameter based on the temporal-motion metric.

5. The method of claim 1 , wherein the one or more calculated video metrics includes an illumination-changes metric.

6. The method of claim 1 , wherein the one or more calculated video metrics includes a blockiness metric.

7. The method of claim 1 , wherein calculating the weighted sum based on the obtained metric-specific weighting factors and the corresponding calculated video metrics includes:
calculating respective products of the obtained metric-specific weighting factors and the corresponding calculated video metrics; and
calculating the weighted sum as the sum of the calculated products.

8. The method of claim 1 , further including:
obtaining training-metric data at least in part by calculating one or more of the video metrics with respect to one or more training sets of video frames;
obtaining training-technique-result data at least in part by applying one or more of the super-resolution techniques in the plurality of super-resolution techniques to one or more of the training sets of video frames; and
calibrating the correlation data based at least in part on the obtained training-metric data and at least in part on the obtained training-technique-result data.

9. The method of claim 8 , wherein calibrating the correlation data includes calibrating one or more of the metric-specific weighting factors.

10. The method of claim 8 , wherein the ranges of weighted sums are defined by respective range boundaries, and wherein calibrating the correlation data includes calibrating one or more of the range boundaries.

11. The method of claim 8 , wherein the obtained training-technique-result data includes one or more face-recognition-system confidence scores.

12. The method of claim 8 , wherein the obtained training-technique-result data includes received user-feedback data.

13. The method of claim 1 , further including performing face recognition on the super-resolved set of video frames.

14. The method of claim 1 , further including:
obtaining a first face-recognition identification and a first face-recognition confidence score for the super-resolved set of video frames;
obtaining a second face-recognition identification and a second face-recognition confidence score for the input set of video frames;
outputting an indication of the first face-recognition identification when the first face-recognition confidence score exceeds the second face-recognition confidence score; and
outputting an indication of the second face-recognition identification when the second face-recognition confidence score exceeds the first face-recognition confidence score.

15. The method of claim 1 , the method further including:
obtaining operational-results data at least in part by performing face recognition on the super-resolved input set of video frames; and
calibrating the correlation data based on one or more of the calculated video metrics and at least in part on the obtained operational-results data.

16. The method of claim 15 , wherein the obtained operational-results data includes one or more face-recognition-system confidence scores.

17. The method of claim 15 , wherein the obtained operational-results data includes received user-feedback data.

18. A system for selecting a super-resolution technique for improved face recognition performance, the system including:
a communication interface;
a processor; and
data storage containing instructions executable by the processor for causing the system to carry out a set of functions, the set of functions including:
calculating one or more video metrics with respect to an input set of video frames;
obtaining a metric-specific weighting factor for each of the calculated video metrics;
calculating a weighted sum based on the obtained metric-specific weighting factors and the corresponding calculated video metrics;
selecting, based on the calculated weighted sum, a particular super-resolution technique from among a plurality of pre-determined super-resolution techniques by referencing correlation data that maps ranges of weighted sums to corresponding super-resolution techniques; and

using the selected super-resolution technique to generate a super-resolved set of video frames by applying the selected super-resolution technique to the input set of video frames.

Assignees

• Motorola Solutions Inc
• Motorola Solutions Inc

Inventors

• Lee Kuan Heng
• O'Connell Kevin J
• Sim Moh Lim
• Zhang Yan

Classifications

• G06V40/173

  face re-identification, e.g. recognising unknown faces across different face tracks · CPC title

• G06V40/167 Primary

  using comparisons between temporally consecutive images · CPC title

• G06V10/993

  Evaluation of the quality of the acquired pattern · CPC title

• G06K9/00295 Primary

  Physics · mapped topic