FORENSIC SKETCH BASED RECOGNITION USING DISCRIMINANT ANALYSIS

Abstract

In this project, we are addressing the problem of matching sketches (forensic) to mug shot images. In previous researches, forensic sketch matching offered only solutions to highly accurate viewed sketches(sketches that are drawn by looking at the person). The difference between forensic sketches and viewed sketches is that the former is drawn by a police artist with the help of the description provided by an eye witness. We here present a framework called local feature based discriminant analysis(LFDA)to differentiate between various forensic drawings. In LFDA we separately express both drawings and pictures using SIFT feature descriptor and multi-scale local binary patterns(MLBP). We then use multiple discriminant projections on subdivided vectors of feature based representation for least separation matching. On comparison to a leading face recognition system, LFDA provides substantial rectification in comparing forensic drawings to corresponding face images.