Performance Analysis of Simulated Annealing Cooling Schedules in the Context of Dense Image Matching

Abstract

Stochastic optimization is widely used in combinatorial computer vision problems, and many variants have been proposed. This contribution aims at analyzing and assessing several flavours of the simulated annealing algorithm. We particularly want to show the optimization performance, convergence speed, and quality of the solution with respect to the algorithm's parameters and cooling schedules. We also verify experimentally that the S.A. algorithm is a global method, i.e., it is able to lock a strong minimum regardless of the initialization. Performance evaluation is conducted in the context of stereo matching.