Advancement Through Research
Biometric is the technology based on biological traits, which exploits the physical and behavioral characteristics of an individual. Ear biometric has gained immense attention over the last years. Because of its consistent shape and rich texture distribution, it is a reliable biometric for human recognition and identification. This paper presents an approach for ear based human identification using Wavelet transformation and Multi-scale Local Binary Pattern (MLBP). It exploits Haar wavelet decomposition up to fourth level and uniform texture distribution over the circular neighborhood region by varying the scale. Two different distance scores are incorporated for classification, namely, match distance and chi-square statistics. The proposed feature extraction and classification method are performed on IIT Delhi Ear Database, which has ear images acquired from 221 different subjects. The experimental results have shown better performance (in terms of accuracy) by an increment in a number of neighbors. The experimental results have shown better performance with the highest accuracy of 97.70% by an increment in a number of neighbors in MLBP, increasing the number of decomposition levels and also using different classifiers.
Biometrics, Wavelet transform, Haar, MLBP, Feature vector, Match distance, Chi-square statistics
