Local Descriptor Approach to Wrist Vein Recognition with DVH-LBP Domain Feature Selection Scheme

Abstract

Local Binary Pattern (LBP) is one of the well-known image recognition descriptors for texture-based images due to its superiority. LBP can represent texture well due to its ability to discriminate and compute efficiency. However, when it is used to describe textures that are barely visible, such as vein images (especially contactless vein), its discrimination ability is reduced, which leads to lower performance. LBP has extensively been implemented for features extraction in recognition system of hand, eye, face, eye, and other images. Nowadays, there are a lot of developments of hand recognition systems as a hand is a part of the body that can be easily used in the recognition process and it is easier to contact the sensor when taking the image (user-friendly). In particular, a hand consists of various parts that can be used, such as palm and fingers. Other parts like dorsal and wrist can also be used as they have unique characteristics, i.e., they are different from each other, and they do not change with ages. Changes in pixel intensity can be derived from skeletal vein images to distinguish individuals in palm vein recognition. In the previous paper, we proposed a method diagonal, vertical, horizontal local binary pattern (DVH-LBP) for implementing the palm vein recognition system successfully. Through this work, we improve our previous procedure and implement the improved method for recognizing wrist. In particular, this study proposes a new and robust directional extraction technique for encoding the functions of the wrist vein in a simple representation of binary numbers. Simulation results show the low equal error rate (ERR) of the proposed technique is 0.012, and the recognition rate is 99.4%.

Hybrid Feature Extractor Using Discrete Wavelet Transform and Histogram of Oriented Gradient on Convolutional-Neural-Network-Based Palm Vein Recognition

“…The process then continues to feature extraction, which can be performed based on two categories of low-level features, namely, structural and textural features. There are several methods for this category of features, such as supervised discriminative sparse principal component analysis neighborhood-preserving embedding (SDSPCA-NPE) [ 31 ], local binary pattern (LBP) [ 32 ], gray-level co-occurrence matrix (GLCM), and histogram of oriented gradient (HOG) [ 33 ]. Based on the results of [ 33 ], HOG showed the best result among the texture features that were extracted [ 33 ] due to its superiority in detecting the degree of differences among transformations and variants [ 34 ], although there are still some reports that low-level features are unrepresentative and unstable [ 35 ].…”