Facial Image Deblurring & Recognition using GFP-GAN
Izaiah Exum, 2nd-year, Computer Science
Dr. Xiaohong Yuan, Computer Science, North Carolina Agricultural and Technical State University
Technology captures facial images of people in a variety of ways, many of which produce blurred images or those of unrecognizable quality. It is important to develop methods to enhance and deblur those images. This research covers the use of the Multi-Frame Labeled Faces Database (MFLFD), the algorithm/method that aims to deblur facial images in the database, and the results and implications of using the method. The database consists of two datasets TRAIN and TEST which contain over 12000 blurred facial images captured from YouTube. This research will discuss what was used to resize the images to 250x250 pixels. The algorithm uses the GFP-GAN framework to deblur and recognize the facial images provided by the database. The GFP-GAN framework uses a degradation removal module and a pre-trained face GAN. By use of the GFP-GAN framework the goal of the algorithm is to remove the features that degrade and blur the image and produce one that is closest to ground-truth. The research will conclude with the challenges and next-steps regarding the algorithm.
Exum, Izaiah, "Facial Image Deblurring & Recognition using GFP-GAN" (2023). Undergraduate Research and Creative Inquiry Symposia. 319.