Font Size: a A A

Research And Implementation Of Speaker Lip Feature Extraction Algorithm

Posted on:2019-05-16Degree:MasterType:Thesis
Country:ChinaCandidate:S S CuiFull Text:PDF
GTID:2428330566464628Subject:EngineeringˇComputer Technology
Abstract/Summary:PDF Full Text Request
In many applications,such as audio-visual speech recognition,facial expression recognition and emotion detection,etc,which are based on intelligent information processing,the detection and extraction of the lip feature plays an important role.The existing lip feature extraction algorithms have many problems,such as large amount of computation,too many manual intervention in processing,and low practicability.After analyzing and comparing the current lip feature information extraction scheme,based on the changing rules of lip movement in the process of pronunciation,a new scheme of speaker lip feature extraction was designed and constructed.Then,the scheme has been discussed and studied the scheme deeply,and the main work contains the following aspects:(1)On the basis of analyzing and comparing the existing lip feature information extraction scheme,a new scheme of speaker lip feature information extraction is designed and a processing model of this scheme is built,which includes sections of video information preprocess,lip contour sequence generation and lip feature information extraction.(2)In the section of video information preprocessing,the preprocess model is designed.And some function modules of the preprocess model have been designed and implemented,which includes sections of video segmentation,face detection and lip region detection.(3)In the section of lip contour sequence generation,an automatic extraction method of lip contour based on YIQ chroma space is designed,so that the lip contour sequence generation model is designed further.And some function modules of the generation model have been designed and implemented,which includes sections of denoising,brightness filtering.(4)In the section of lip feature information extraction,the changing process of lip type can also be regarded as the serialization of the set of pronunciation video frames.Therefore,the value of speaker lip feature can be obtained through the boundary distribution characteristics and change sequence of the lip contour.Then,the model of lip feature information extraction is designed.And some function modules of the extraction model have been designed and implemented,which includes sections of key frame selection,feature data acquisition.(5)Based on the application of monosyllabic recognition,the proposed lip feature extraction scheme in this paper has been testified,which has proved that the scheme has good accessibility and practicability.
Keywords/Search Tags:Video, Speaker, Lip Feature, YIQ Chroma Space, Frame Screening
PDF Full Text Request
Related items