Lip Feature Extraction And Mouth-shapes Classification Research Based On Image

Recently, computer lipreading technology has received the concern of more andmore researchers, which is regarded as an auxiliary method for Automatic SpeechRecognition (ASR). Lipreading research mainly includes: the basic characteristics ofthe mouth figure extraction, description and said, as well as lip figure recognition andunderstanding, it is the comprehensive understanding of the relevant information fromnumber of channels.This paper mainly studies methods of the face location, lip area location, lipfeature extraction and Mouth-shapes classification in a human face image. Firstly,based on analysis and compares of the current mainstream method of lipreadingdeeply, illumination compensation is performed based on the histogram equalization.Then, we constructed a skin-color model for skin detection in YCbCr color space sothat the approximate face region can be obtained. Next, Fisher transformationapproach is used to strengthen the gradient information of the lip region and then thelip region can be obtained based on the geometry features of human face.Secondly, Based on the separability of different components in different colorspaces for lip and non-lip, a lip segmentation method based on ‘a’ component and ‘L’component of LAB color space is used to split outer lip and inner lip. This methodperformance very good and it can create the threshold automatically. Then severalcurves based on Deformable Templates are used to make up a mathematics formularyof the lips. We proposed a method to improve match result for get the shape feature oflips.Finally, Mouth-shapes in the face image are classified into four types by thecharacteristics of Chinese vowel Lamiaceae: a-style, o-style, e-style and i-style.Experimental results show that our method is effective and meaningful in practice.