Fingerphoto deblurring using deep learning GAN architectures

Patent metadata

Field	Value
Publication number	US-12555207-B2
Application number	US-202318112701-A
Country	US
Kind code	B2
Filing date	Feb 22, 2023
Priority date	Feb 22, 2022
Publication date	Feb 17, 2026
Grant date	Feb 17, 2026

Abstract

Various examples are provided related to fingerphoto deblurring. In one example, a method includes generating, using a guided-attention (GA) mechanism, an intermediate feature map of a blurred image of a fingerprint and generating a deblurred image of the fingerprint based at least in part upon the intermediate feature map. The GA mechanism can generate the intermediate feature map by generating an attended feature map from an input feature map based upon a predicted attention map and adding the input feature map to the attended feature map. A system can include processing circuitry and a fingerphoto deblurring application that, when executed by the processing circuitry, causes the processing circuitry to generate the intermediate feature map and generate the deblurred image.

First claim

Therefore, at least the following is claimed:

1 . A method, comprising:
obtaining a blurred image of a fingerprint; generating, using a guided-attention (GA) mechanism, an intermediate feature map of the blurred image, wherein the GA mechanism generates the intermediate feature map by:
generating an attended feature map from an input feature map based upon a predicted attention map; and
adding the input feature map to the attended feature map; and
generating a deblurred image of the fingerprint based at least in part upon the intermediate feature map.

2 . The method of claim 1 , wherein the predicted attention map is generated from the input feature map using a convolutional layer and a Sigmoid function.

3 . The method of claim 1 , wherein generating the attended feature map comprises multiplying the input feature map with the predicted attention map.

4 . The method of claim 3 , wherein the attended feature map is normalized by a 2D batch normalization.

5 . The method of claim 1 , wherein the input feature map is generated from an intermediate feature map generated by a preceding GA mechanism.

6 . The method of claim 5 , wherein the input feature map is generated by up-sampling the intermediate feature map generated by the preceding GA mechanism.

7 . The method of claim 1 , wherein the deblurred image is generated from the intermediate feature map using a convolutional layer.

8 . The method of claim 1 , wherein the deblurred image is a deblurred output image of the fingerprint having a resolution equal to the blurred image of the fingerprint, the deblurred output image generated by:
up-sampling the intermediate feature map; and applying a convolutional layer.

9 . The method of claim 1 , comprising identifying a blurring type associated with the blurred image based upon the intermediate feature map.

10 . The method of claim 1 , comprising generating a second intermediate feature map of the blurred image using a second GA mechanism, the second intermediate feature map having a resolution less than the intermediate feature map.

11 . The method of claim 1 , comprising determining a reconstruction loss based upon a comparison of the deblurred image and a corresponding ground truth image.

12 . A system, comprising:
processing circuitry comprising a processor and memory; and a fingerphoto deblurring application executable by the processing circuitry, where execution of the fingerphoto deblurring application causes the processing circuitry to:
generate, using a guided-attention (GA) mechanism, an intermediate feature map of a blurred image of a fingerprint, wherein the GA mechanism generates the intermediate feature map by:
generating an attended feature map from an input feature map based upon a predicted attention map; and
adding the input feature map to the attended feature map; and

generate a deblurred image of the fingerprint based at least in part upon the intermediate feature map.

13 . The system of claim 12 , wherein the predicted attention map is generated from the input feature map using a convolutional layer and a Sigmoid function.

14 . The system of claim 12 , wherein generating the attended feature map comprises multiplying the input feature map with the predicted attention map.

15 . The system of claim 12 , wherein the input feature map is generated from an intermediate feature map generated by a preceding GA mechanism.

16 . The system of claim 12 , wherein the deblurred image is generated from the intermediate feature map using a convolutional layer.

17 . The system of claim 12 , wherein the deblurred image is generated by:
up-sampling the intermediate feature map; and applying a convolutional layer.

18 . The system of claim 12 , comprising execution of the fingerphoto deblurring application causes the processing circuitry to generate a second intermediate feature map of the blurred image using a second GA mechanism, the second intermediate feature map having a resolution less than the intermediate feature map.

Assignees

• West Virginia Univ Board Of Governors On Behalf Of West Virginia Univ

Inventors

• Nasrabadi Nasser M
• Dawson Jeremy M
• Joshi Amol
• Dabouei Ali

Classifications

• G06T3/4046

  using neural networks · CPC title

• G06T2207/20016

  Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform · CPC title

• G06T5/60

  using machine learning, e.g. neural networks · CPC title

• G06V10/82

  using neural networks · CPC title

• G06V40/1347

  Preprocessing; Feature extraction · CPC title