Lip feature extraction based on improved jumping-snake model
Xinjun Ma, Yan Long, Qianyuan Zhong · 2016
A novel feature extraction method used in computer lip-reading is proposed, which attains lip parameter based on lip-points detection. Then, the division-detection strategy is presented, which firstly uses face template matching method to focus on lip region, and secondly uses the improved jumping-snake method to detect lip-points. The improved jumping-snake method provides new ways to construct gradient field, and to update seed. Moreover, the error detection-recovery strategy is added to enhance the performance of jumping-snake. In the final, effectiveness and stability of the proposed algorithm is proved by some experiments.