Particle Swarm Optimization for the Fusion of Thermal and Visible Descriptors in Face Recognition Systems

In this paper, we propose a novel face recognition system based on fusing thermal and visible descriptors. The proposed approach is divided in two steps: training and validation. In the training stage, the system obtained the optimal weights from the particle swarm optimization (PSO) algorithm to maximize the recognition rates obtained from different combinations of local descriptor methods using a standard thermal face database (Equinox database). The weights were then used to fuse visible and thermal face descriptors to achieve high recognition rates during the validation stage using the Pontificia Universidad Católica de Valparaiso-Visible Thermal Face (PUCV-VTF) database. Three local matching methods were used to perform the face recognition: local binary pattern, histograms of the oriented gradients, and local derivative pattern. In addition, this paper considers a comparison with the following methods: a previous work based on Genetic Algorithms and a modified PSO approach. The results of this paper show recognition rates over 99% for the PUCV-VTF database, largely surpassing the results for Genetic Algorithms. The fusion methodology is found to be unaffected to variations in illumination and expression conditions, combining the visible and thermal information efficiently through the PSO algorithm, and thus choosing the optimal regions where a given spectrum is more relevant.