Comprehensive survey of face super-resolution techniques

Tomar A.S.; Arya K.V.; Rajput S.S.; Rodriguez C.R. · Digital Image Enhancement and Reconstruction (2022) · DOI: 10.1016/b978-0-32-398370-9.00016-0

Resumen

Face hallucination is one of the domains of image super-resolution, where super-resolution takes the place of face images. It is an attractive domain of research in the field of digital image processing due to its wide application in various fields, such as surveillance, face recognition, facial expression recognition, etc. In the real-world scenario, captured face images are noisy or small in size due to low-resolution imaging system, noise introduced by the image sensors or motion between the object and sensor. These captured face images are low-resolution images that contain less information, poor quality of features; hence it is very difficult to identify the person and their facial expression. The face super-resolution technique uses upsampling and feature improvement to construct the high-resolution or super-resolution image. This chapter provides a detailed survey of learning-based face hallucination methods and a detailed description of Bayesian, regularization, subspace learning, and deep learning methods. Issues and challenges with the existing super-resolution methods are discussed to highlight the problems and future direction to solve them, which will help the researcher to develop the new face hallucination methods.

Instituciones: Universidad Nacional Mayor de San Marcos