PROJECT TITLE :

Affine-Transformation Parameters Regression for Face Alignment

ABSTRACT:

Face alignment is a vital method in facial analysis. Cascaded linear regression approaches have shown the potential to achieve the state-of-the-art accuracy on varied face alignment datasets. But, most of these approaches only learn to map coordinate offsets of the key points from image features. This regression strategy will be simply trapped in local optima. We have a tendency to propose a unique regression strategy by introducing affine transformation. First, the most effective affine-transformation parameters between the initial mean form and the bottom truth are estimated by Procrustes analysis. Subsequently, we have a tendency to base the mapping from image options on the best affine-transformation parameters. Experimental results indicate that this strategy will scale back the offsets between 2 shapes significantly. Combined with coordinate-offset regression strategy, the hybrid approach produces a remarkably performance in term of accuracy, coaching time, prediction rate, and therefore the model size. Moreover, the affine-transformation parameter regression strategy will be considered as a form-initialization technique which will be combined with other initial form-based mostly face alignment algorithms to enhance the face alignment accuracy.