The Research Of Static Lip Recognition Based On HSV Transformation

In daily life, communicating and learning, human perception of language is made of a plurality of channels from the confluence. Either obtain the corresponding voice information through voice channel, or get the lip movement information through visual observation to assist in the understanding of speech information. Thus, it is very important to capture information on the lip to lip recognition research, not only can improve the accuracy rate of voice recognition in the noise and interference environment, can solve the deaf communicate problems,but also can be applied to many areas such as multimedia systems, intelligent human-computer interaction and identification system. In recent years, through the realization of human-computer interaction technology which is automatically lip-reading recognition has been widespread attention, it has become a new area of research. Most research studies scholars combine speech recognition with lip-reding to achieve the desired higher recognition rate and accuracy. Furthermore, in previous studies, the speech recognition technology are advancing rapidly reaching maturity, has far exceeded the lip recognition technology. Therefore, the main purpose of this paper is to research the key technology of lip recognition process, and establish an lip recognition system based on isolated word, lay the foundation to communicate with the deaf combined with gesture recognition.In order to achieve lip recognition, the first condition is a lip accurate positioning,the effect of positioning plays a important role in the accurate of lip reading. Based on the face recognition and positioning the eye in order to gain full pixel area of the lips,and then on the lip feature extraction and processing, finally to isolated word recognition purpose. In the face recognition stage, we use the color segmentation method and compare the the advantages and disadvantages of various common color models. We find the RGB and HSV color model is sensitive to the red zone and YCbCr color model is weak in edge detection, easy to mistake the skin color area as a non-color pixel, while the combination of HSV and YCbCr color models can getbetter results, avoid the effects of the red zone and light intensity. In the human eye positioning stage, firstly, we can quickly get the eye candidate region in the scope of the face by the pixel value of the gradation integration method, and then scan the candidate region to locate the eye area with geometric features like a circular, which is based on the Hough transform circle detection algorithm. In the lip extraction stage,the traditional HSV color space is improved and transformed to enhance the contrast between lip area and surrounding pixels, so the effect of lip segmentation is more ideal. In the lip recognition stage, firstly, the lip geometry features should be extracted,and each feature should be function testing and weighted, then use the weighted feature to match with the template library baesd on the dynamic time warping algorithm(DTW), so as to achieve the purpose of lip recognition.This paper deeply discuss the algorithms and techniques in each module of lip reading, propose a static recognition algorithm based on the existing algorithm which has been optimized and improved, greatly improved the recognition rate and effectiveness of the system. Experiment results show that the designed lip recognition system is tested to be reach 83% recognition rate in the isolated word recognition. But there is still a certain distance with the dynamic lip recognition technology, we need to make continuous efforts to achieve better human-computer interaction in the future.