Font Size: a A A

Research On Lip Reading Method Based On Hidden Markov Model

Posted on:2019-07-03Degree:MasterType:Thesis
Country:ChinaCandidate:Y CaiFull Text:PDF
GTID:2428330545496024Subject:Electronic and communication engineering
Abstract/Summary:PDF Full Text Request
With the rapid development of human-computer interaction,speech recognition,as one of the most efficient and convenient human-computer interaction methods,has received extensive attention.But in the complex environment,the speech recognition by a lot of interference,can not meet the normal needs of human.Therefore,lip reading begins to rise,this technology has limitless foreground in the field of intelligent human-machine interaction,video big data mining,and video monitoring etc.This paper understood the lips of discourse content by the visual features of the speakers and mainly researched on the lip area location,the feature extraction and lip reading.Lip area location is the basis of lip reading system.We choose the more mature AdaBoost algorithm for face detection and location.We get the ROI(region of interest),by the fixed position of lip in the face.The lip feature extraction is the key to the lip reading system.This paper gets the key point of lip contour for curve fitting.The curve parameters are as the shape feature of lips.Then using curve fitting to establish the lip contour model,obtaining visual geometric features.A combination of both form the overall visual feature vector.The optimal locally sensitive discriminant analysis(OLSDA)is used to reduce the dimension of feature vector operation,keep the main features of information,reduce redundancy.Lip reading is the determinant of the lip language recognition system.This paper is based on Hidden Markov Model,to quantify the clustering of lip sequence feature vector with K-means clustering algorithm,as the input of training and reading.Considering the demand of this paper,we use the home-made video database as a sample for the experiment.The experimental results show that the lip reading recognition system in this paper has good performance,and proves the feasibility of the proposed lip localization algorithm,the feature extraction and the algorithm of reading.
Keywords/Search Tags:lip area location, feature extraction, lip reading, Hidden Markov Model
PDF Full Text Request
Related items