Improved active shape model for variable illumination conditions

Automatic facial landmarking is an important problem in face recognition, identification and classification applications. Active Shape Model (ASM) and its derivatives have become increasingly popular and used intensively for automatic facial landmarking. However, since their performances are highly dependent on initialization, illumination conditions and pose, starting with a poor initial shape or testing images with bad lighting can drastically affect the results. In this study, we propose some improvements on ASM for better face alignment under variable lightning conditions. Our improvements include remedies to provide accurate fitting results: (i) using location, scale and rotation information derived from eye pupils, (ii) utilizing a multi-directional profiling scheme and, (iii) neutralizing images to overcome the effects of variable illumination conditions. Results of our proposed method are compared to traditional ASM and another popularly used method called Stacked ASM (STASM). Experimental results on homogenously illuminated images show that our method has better performance compared to the classical ASM and exhibits similar performance to STASM. On the other hand, for non-homogenously illuminated images, our method outperforms STASM.