EAR RECOGNITION USING LOCAL BINARY PATTERN

Abstract

In this paper an efficient complete Ear Recognition System (ERS) has been proposed based on the Local Binary Pattern (LBP) approach that can investigate maximum recognition rate; hence it can be used for surveillance applications. The feature extraction is based on calculating the LBP feature for the ear image and dividing the resultant LBP image into several overlap regions, and then extracts the histogram from each region. These histograms are considered as a similarity measure in the classification phase. To evaluate the proposed approach, the Indian Institute of Technology (IIT) Delhi processed ear image dataset has been considered, which contains 125 individual, each with at least three images acquired in the age group between 14 and 58 years. Practical experiments are employed on the proposed ERS to find the best image division regions at best LBP parameters (radius and neighbors) that lead to maximum recognition rate. Detailed experiments show that the proposed system achieved 93.75 % rank-one recognition rate. Furthermore, an experimental study is achieved to examine the less Equal Error Rate (EER). Some identities from the database are considered as imposters. In a verification scenario, the system achieved an Equal Error Rate (EER) of 14.94 %. The Receiver Operating Characteristics (ROC) curve showed that the Genuine Acceptance Rate (GAR) is about 84%.