Recognition of hand-gestures using improved local binary pattern

This paper presents an improved LBP (local binary pattern) texture descriptor, which is used to classify static hand-gesture images. The descriptor makes full use of the correlation and the differences of pixel gray value in the local regions, simple and fast coding; the descriptor is a highly discriminative texture operator, and can represent different characteristics of static hand-gesture images. Initial, we introduced the texture descriptor of improved LBP. Then, we extracted features of hand-gesture images by using improved LBP descriptor and uniform patterns encoding. At last, an Adaboost classifier performed the hand-gesture recognition task. A dataset with many hand-gestures (12 types, 600 hand-gesture images) was built, including some large pose-angle hand-gesture images. The experiment results show that the prominent performance of improved LBP descriptor; the capability of is close to LBP, superior to LAP (local angular phase). Meanwhile, the descriptor is robust to nonlinear illumination, scaling.