Research On Lip Visual Feature Extraction And Speech Recognition

In today's society,with the advent of computer and the gradual popularization,people's demand for information shows a growth of geometric progression,one of the main ways of information exchange is through voice communication.In the noisy environment,people's perception of voice information is interferenced and significantly reduced,at this time voice communication seems a bit stretched.In recent years,the rise of image processing and pattern recognition technology has made the computer vision technology been widely concerned by researchers.After a lot of exploration and analysis,it is found that lip and its dynamic characteristics play an important role in the process of human perception of language.The speaker's lip movement can make the audiences understand or partially understand the content of his/her speech.Using the lip visual information to speak the language at the present stage and the future period of time has a very important theoretical significance and wide application prospects.The research content of this paper is based on lip segmentation,focuses on the lip visual feature extraction and speech recognition.In this paper,we propose an improved active contour model(ACM)to realize the precise segmentation of the lip contours.Before that,it is also necessary to make a rough positioning for the lip region image to reduce the time complexity of the algorithm.In this paper,the face is detected and positioned by applying OpenCV technology.By locating the position of the human face,the rough region of the lip,that is,the region of interest(ROI),is calculated.In this paper,we also analyze the effectiveness of the initial contour for the active contour model,and propose a method to automatically reduce the number of iterations and improve the efficiency of segmentation according to the lip deformation.In the aspect of visual feature extraction,this paper tends to use the segmented lips contour to determine the key points on a series of contours.Then,through the fitting of these key points,the curve of the lip contour is determined,the parameters of the curve are taken as visual features.Integrated with the geometric features,there forms the formal eigenvector for recognition input,which is considered as a prerequisite for speech recognition.Finally,in the aspect of recognition algorithm,this paper chooses a relatively classic and stable classifier,named support vector machine.Through some experiments,it is proved that it has a good performance in speech recognition and achieves satisfactory recognition result.