ABSTRACT
Image super-resolution is a topic of great interest. It has significant
applications in ultra-high-definition TV display, image resizing, face
recognition, object recognition, video coding and surveillance. The
objective is to construct a high-resolution image from one or
several low-resolution images, while minimizing visual artifacts.
Classical approaches have come to a quality limit because of the
constraint on manual filter design and limited design structures.
Learning approaches allow super-resolution algorithm design to
be adaptive to training data and automatically form thousands of
filters/adapters for the best super-resolution. In this chapter, wewill
introduce (i) some conventional learning approaches, (ii) random
forests, and (iii) Convolutional Neural Network (CNN) for effective
image super-resolution.