Efficient Integrated Model for Feature Descriptor and Texture Classifier using WDOLBP Histogram Forhand-Dorsa Vein Recognition System

  • C Premavathi Part-Time Ph.D Scholar, Periyar University,Salem, Tamil Nadu, India, Assistant Professor,Department of Computer Science, Navarasam Arts and Science College for Women, Arachalur, Tamil Nadu, India
  • P Thangaraj Professor and Head, Department of Computer Science and Engineering, Bannari Amman Institute of Technology, Sathyamangalam, Tamil Nadu, India
Keywords: Weber Local Binary Pattern, Pattern Recognition, Feature Descriptor, Feature Rduction, Classification


Hand-dorsa Vein Recognition System is a biometricauthentication system using inherentphysiological characteristics to enable identification of individuals. In this paper a new integratedframework has been proposed to specify the features of hand vein images and identify the individualimages with classification method. The framework consists of three primary components calledFeatureExtraction, Dimensionality Reduction, and Texture Classification and each have subcomponents.The feature extraction component is based on Weber Differential OrientationLocalBinary Pattern (WDOLBP) where as the feature reduction is based on Principal Component Analysis(PCA) and the texture classification component is based on nearest neighbour classification. For eachpixel of the input image, WDOLBP descriptor is computed with two features called differentialOrientation (DO) and Local Binary Pattern (LBP). By combining DO and LBP features (calledWDOLBP feature) per pixel, the feature vector is represented in a histogram, which is called WDOLBPhistogram. Feature reduction component is required to reduce the dimensionality of feature images ina histogram. In this work, PCA is suggested to reduce the features of images which then can be used inClassification. The final component is a classifier is based nearest neighbour classification method.The proposed method is evaluated on a NCUT Dataset contains 2040 images from Prof. Yiding Wang,North China University of technology (NCUT) (Wang et al, 2010). Similarity measures of variousclassification methods such as Chi-square, Cityblock, Euclidean, Chebychev and Minkowski arecomputed and compared for the better performance. The experimental results show that the proposedintegrated framework performs better than other feature representation

Abstract views: 316 times
PDF downloads: 0 times