ProDeblurGAN: Progressive growing of GANs for blind motion deblurring in face recognition

3Citations
Citations of this article
1Readers
Mendeley users who have this article in their library.
Get full text

Abstract

Elimination of motion blur is one of the key challenges in face recognition, especially from the video feed of low quality surveillance cameras. We present ProDeblurGAN, a progressively growing Generative Adversarial Nets (GAN) for single photograph motion deblurring. The current state-of-the-art DeblurGAN network performs well in structural similarity and visual appearance. However, the discriminator’s inability in distinguishing real and fake images concerning to finer or lower-level details result in the generator synthesizing low quality deblurred image. With recent advances in the progressive growing of GANs, we propose a motion deblurring model to utilize the advantages of such progressive adversarial training and generate a high quality deblurred image. The approach to gradually increase layers in training instead of initializing all at once facilitates the generation of higher resolution images. Also, at higher resolution, the finer details like the facial hair, moles, freckles and so on, are considered as features and used in the learning process. The proposed generator takes the blurred image as input as opposed to random noise in the case of traditional GAN networks. Implementation of the proposed method in face recognition systems will enhance the quality of the captured image and the accuracy of recognition of the face. The quality of the deblurred image generated by ProDeblurGAN is evaluated using state-of-the-art metrics.

Cite

CITATION STYLE

APA

Mahalingaiah, K., & Matichuk, B. (2020). ProDeblurGAN: Progressive growing of GANs for blind motion deblurring in face recognition. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 12015 LNCS, pp. 439–450). Springer. https://doi.org/10.1007/978-3-030-54407-2_37

Register to see more suggestions

Mendeley helps you to discover research relevant for your work.

Already have an account?

Save time finding and organizing research with Mendeley

Sign up for free