Face Recognition Using Sparse Representation

Abstract

Face recognition became most important aspect in daily life. It has many application including biometrics which used on daily bases. For authentication of many devices we need to take help of face recognition.

In this thesis, the face recognition is done by the Sparse representation classifier. Based on dictionary size and various application thesis has divided into three different chapters-SRC,ESRC, FDDL. face recognition method uses stored database as dictionary which consist of frontal images of different persons. This images may contain variation in expression, different light illumination and it may contain occlusion like hat ,scarf and sunglasses. Some images does not belongs from of any training or testing dataset. These kind of problems may occur. Though the kind of problems may occur algorithm should work robustly to give most accurate face recognition rate.

In SRC large number of training data set is used as single dictionary. For extracting the information different feature extraction method are applied like down sampling. SRC algorithm usesl1norm to calculate sparse coefficient. Residual had been calculated to classify the test image correctly.

The application where few or only single training image required as dictionary image. In such cases ESRC has used. Intra class variant dictionary has made along with dictionary. So as to achieve better face recognition rate for under sampled images.

FDDL uses sub-dictionary for each class and discriminative sparse coding coefficient to achieve better results than SRC.