Hand gesture recognition based on HOG-LBP feature

With the rapid development of information technology, human-computer interaction (HCI) is now experiencing the transition from traditional command line interface to novel natural user interface such as speech and gesture, thus vision-based hand gesture recognition is one of the key technologies to realize natural HCI. However, the performance of gesture recognition is often influenced by variations among lighting conditions, complex backgrounds and so on. This paper proposes a new fusion approach of hand gesture recognition by combining the HOG and uniform LBP feature on blocks, in which HOG features depict hand shape and LBP features depict hand texture. Support Vector Machine with radial basis function (RBF) as kernel function is adopted to train the hand gesture classifier. Experimental results show that HOG-LBP fused feature performs well on two sub-datasets from NUS hand posture dataset-II, reaching a relative high recognition accuracy of 97.8% and 95.07% respectively. The comparison experiments among HOG-LBP, HOG and LBP features also show that the HOG-LBP feature performs better than one single feature.