Face Alignment using Boosted Appeareance Model (Discriminative Appearance Model)

Abstract: This thesis explores decriminative face alignment using Boosted Appearance Model (BAM). In this method face alignment is done by maximizing the score of the trained two classifier which learns both correct and incorrect alignment and is able to distinguish correct and incorrect alignment so that the correct alignment gets maximum positve score. During the training stage we trained Point Distribution Model (PDM) which acts as shape model and a boosting based classifier based on Haar like Rectangular Features which we call as Boosted Appearance Model (BAM). This algorithm iteratively updates the shape parameters of the PDM by the well known optimization method known as gradient ascent, such that the classification score is maximized. When we test our algorithm on the images the initial parameters will likely have a negative score and these parameters are updated so that the final parameters will have the maximum positive score this will indicates alignment is finalized. The main applications are tracking, Medical Image Interpretation, Industrial Inspection etc.