Daniel Otero, PhD Candidate
Department of Applied Mathematics, University of Waterloo
As is known, many imaging tasks can be carried out by solving an optimization problem. In these problems, the square of the Euclidian distance is widely used as a fitting term since it is convex, differentiable, and mathematically tractable; however, it has been shown that this metric is not the best choice when it comes to measure similarity between images. To overcome this difficulty, some researchers in the image processing community have employed the structural similarity index measure (SSIM) as a fidelity term to carry out a variety of SSIM-based imaging tasks. In this talk, we present a general framework that encompasses the different types of SSIM-based optimization problems that can be found in the literature, as well as SSIM-based imaging tasks that had not been addressed before. Also, we discuss several algorithms that we have introduced to solve such SSIM-based optimization problems. Applications and results are also presented.