Improving hand vein recognition by score weighted fusion of wavelet-domain multi-radius local binary patterns

Abstract

Among biometric modalities, hand vein patterns are seen as providing an attractive method for high-level security access applications owing to high impenetrability and good user convenience. For biometric recognition based on near-infrared dorsal hand vein images, Local Binary Patterns (LBP) have emerged as a highly effective descriptor of local image texture with high recognition performance reported. In this paper, the traditional approach with LBP applied in the spatial domain is extended to multi-radius LBP in the wavelet domain to provide a more comprehensive set of feature categories to capture grey-level variation characteristics of vein patterns, and score weighted fusion based on the relative discriminative power of each feature category is proposed to achieve higher recognition performance. The proposed methodology is shown to provide a more robust performance with a recognition rate in excess of 99% and an equal error rate significantly less than 2%.

Improved multiscale local phase quantisation histogram for the recognition of blurred dorsal hand vein images

“…2) were randomly chosen as the training set. A detailed comparison was conducted with the local binary pattern (LBP) [13], histogram of oriented gradient (HOG) [14], Weber local descriptor (WLD) [15], scale-invariant feature transform (SIFT) [16,17], Gabor [18], LPQ [8], and MLPQ [6], which are state-of-the-art algorithms in dorsal hand vein recognition. The correlation coefficient between adjacent pixels ρ was 0.99; the sizes of local neighbours M were 23 × 23, 25 × 25, and 27 × 27, respectively; the values of the corresponding frequency parameter a were 1/23, 1/25, and 1/27, respectively.…”