High precision PSO and FLS integrated method for facial landmark localization

Abstract

Automatic finding exact location of facial salient points under translation, rotation and changing lightning illumination is a considerable task in face image processing. This paper presents a multistage procedure for finding landmark points on a rigid object like human face. Gabor filter jets make EBGM, very effective but computationally expensive. In proposed method, searching landmark points using Gabor filter jets is optimized by using particle swarm optimization (PSO) and similarity between model jet and extracted jet as cost function. After locating first landmark, the location of next landmark is estimated and then is refined by local search criteria (FLS) until localizing of all desired 5 landmarks. Model jets are used for accounting pixels and can be extracted manually from landmark points of same identity for more robustness and accuracy. Results based on the proposed approach are included to prove the accuracy and low computational cost of proposed method comparing the exhaustive search.