LOW RESOLUTION FACE RECOGNITION USING DEEP LEARNING
Date
2023-05
Journal Title
Journal ISSN
Volume Title
Publisher
I.O.E. Pulchowk Campus
Abstract
In recent years, face recognition systems have achieved impressive performance and results
using variety of algorithms and methods but such methods often fail to recognize a face image
of low resolutiom. Face Recognition(FR) degrades when faces are of very low resolution
since many details about the difference between one person and another can only be captured
in images of sufficient resolution. In order to have better face recognition in low resolution
environment, our project uses a Convolutional Neural Network(CNN) model of a Residual
Network Architecture to reconstruct the low resolution image into a higher resolution image
and another CNN model to extract features from the newly reconstructed image to compare
and recognize the face using a classifier. In case of the project, the low resolution image is
taken of resolution 32X32 and the higher resolution image is of 128X128. The PSNR value
for the super resolution model is 29.3256 dB and SSIM value is 0.7686. The accuracy of the
Face Recognition model is 86.32% The performance of proposed method is evaluated on a
custom face dataset using confusion matrix and it shows a decent precision and recall values.
Description
In many surveillance scenarios in real life, people may be far from the camera and their faces
may be small in the field of view resulting in low resolution images. Such low resolution
images can seriously degrade the performance of conventional face recognition systems which
have been mainly developed for recognizing high quality images in controlled conditions.
Keywords
Low Resolution(LR),, High resolution(HR),, Super Resolution,