Application of Deep Learning in Super-resolution Processing of Face Images

Abstract

The resolution of an image is generally defined as the number of pixels contained in a unit size. The higher the resolution, the more details it contains and the clearer the image is. Because the results of reconstructing high-resolution images from low-resolution images are not unique, the super-resolution of images is a morbid inverse problem, and it is also a challenging and open research topic in the field of computer vision. With the development of machine learning and deep neural network, this paper studies the application of DL (Deep Learning) in face image super-resolution processing. The results show that the algorithm in this paper combines detail enhancement module and synthesizes fine-grained structure from high-resolution example image, which can generate low-frequency details of the image and transmit high-frequency details from the example image to the basic image for enhancement. It can be seen that the algorithm in this paper is better than other methods in evaluation index and visual effect. However, there are still some shortcomings in this algorithm and experiment, which need further research and improvement. The two image super-resolution algorithms proposed in this paper are both aimed at improving the perceptual quality of reconstructed images.