Hand gesture recognition using saliency and histogram intersection kernel based sparse representation

Nowadays, sparse representation classification (SRC) has been widely applied in various computer vision areas such as face recognition. However, few researchers have applied SRC in static hand gesture recognition. In our paper, we propose to employ saliency based feature and sparse representation for hand gesture recognition and make in-depth researches in sparsity term parameter and sparse coefficient computation. In addition, literatures show that SRC can not deal with non-linear features well and may produce bad recognition results, so we propose to employ histogram intersection kernel function to map the original features into kernel feature space and use sparse representation classification in the kernel feature space. Furthermore, we compare SR with Support Vector Machine (SVM), Artificial Neural Networks (ANN), Bayesian Network (BN) and Decision Tree (DT). At last, experimental results show that the recognition rate obtained using l1ls_featuresign algorithm has a higher recognition rate than that of l1_ls algorithm and Sparse Representation outperforms all other classifiers compared. In addition, the performance comparison on different kernel functions and different features is also conducted. The average recognition rate of saliency based feature on histogram intersection kernel is 98.91 %, indicating the effectiveness of the proposed saliency based feature and the histogram intersection kernel.