Hand Gesture Recognition with Color Descriptors

The recognition of hand gesture based on computer vision is still a challenging problem, due to the high degrees of freedom or variability in the characteristics of hand gesture with respect to individuals. In this paper, combined with the color descriptors based on scale invariant feature transform features, a vision-based recognition with class specific DPL algorithm, or dictionary pairs learning, is proposed for the applications of hand gesture recognition, which may be considered as the extension of over complete dictionaries learning based on sparse representation. And it is also unnecessary to obtain fixed-length feature vectors for each hand image in the DPL related classification. The proposed approach has been evaluated on a publicly available hand gesture dataset of American sign languages with full digits and alphabet. The obtained experimental results indicate that the provided approach successfully recognizes 36 classes of hand gesture. In another way, the related experiments state that the proposed recognition scheme can offer high recognition accuracy with 97.18% for isolated hand signs and robust to noisy data. Then, it has been verified that DPL algorithm based on color descriptors can be applied to hand gesture recognition.