Features Extraction And Recognition Of Lip-reading Based On The Inner Lip

Audio speech recognition (ASR) technology has developed more and more mature recently in laboratory conditions. As important supplementary information between people's communication, lip-reading recognition technology improves speech recognition rate in the noisy environment. Meanwhile, it plays an inestimable role on accessing special channels to obtain information. Lip-reading recognition technology involves in several areas, such as pattern recognition, computer vision, image processing, and etc. Using computer to achieve the technology has far-reaching impact to the man-machine communication technology. The research about lip-reading recognition technology starts late and is a new subject in our country, more and more researchers focus on it.Based on the extensive references at home and abroad and doing research and comparative analysis on lip-reading recognition technology, especially on the aspects of lip location, features extraction and recognition. The paper designs a system of features extraction and recognition of lip-reading based on inner lip. The inner lip is stressed to be the central role by the paper.For lip location, the paper proposes a method of average gray lip, build a average gray lip template by the lip-reading database and match the precise position of lip region. The method is simply operate, it can locate lip district correctly and the percent is above 90%.For feature extraction, the inner lip is stressed to playing the central role in the whole process. The paper builds the inner lip model, and redefine the expression of curve function of inner lip, which is more sensitive, the geometry and gray features of inner lip is operated as the most important features in lip-reading. To the existing double lip lines model, the new lip model based on the inside lip is simplified and optimized. The method can extract main features effectively, in the meanwhile, it reduces computation complexity and has high recognition. For recognition, the paper proposed a method of features clustering and recognition, use clustering algorithm to obtain the center of geometric and gray features of training images, and then use the distance function to finish the recognition of images. The clustering algorithm applied in recognition of lip-reading is a new trying, and the method can achieve the recognition processing.In this system, the key point of the research is the inner lip model construction and application, this is a new trying. The experiment gathers visual information of single pronunciation of several voices, use K-means clustering algorithm and different distance function rules to finish the recognition process. Results show that this method greatly reduce the amount and the complexity of computation and have high recognition, the new trying is feasible.