Enhanced Local Ternary Pattern method for Face Recognition

Abstract

Biometrics is a term used to determine an individual's identification based on physiological or behavioral traits. Such physiological or behavioral characteristics differ from person to person. For this reason, it is more secure and popular to authenticate the person using biological characteristics than other conventional authentication methods. The Local Binary Pattern (LBP) face recognition system is widely used but is noise sensitive. For the purpose of improving the performance, a descriptor of a local texture called Local Ternary Pattern (LTP) is introduced, which is more discriminating in uniform regions and less noise sensitive. The proposed method called Enhanced LTP (ELTP), uses pre-processing technique. Here, the input image is pre-processed using Gamma Correction and Histogram Equalization. The LTP is applied on pre-processed image to get the finalized feature vectors. Experimentation is conducted on the standard datasets ORL, UMIST and VTU (VISA) face datasets. It is proved that ELTP shows better accuracy than other face recognition methods.