Abstract

Because high-quality facial images have an effect on face detection and recognition areas,the quality improvements of the poor facial areas are essential to raise the accuracy of those areas from the low-resolution(LR) images. Recently, convolutional neural network(CNN) based Super Resolution(SR) methods have shown the visual quality improvements to generate high-resolution(HR) images from LR images. In this paper, we propose dual branch based super resolution method for the quality enhancement of facial areas. The proposed method consists of two parallel CNN networks to focus on the facial areas. Experimental results shows the significant quality enhancement, compared to the existing SR networks

Highlights