Research On The Algorithm Of Lip Segmentation Based On Active Contour Model

In today's rapid development of human-computer interaction,speech recognition,as an important way of human-computer interaction,often effected on performance by noisy environments.Therefore,many scholars put the research focus on the visual lip reading technology to try to improve the defect of speech recognition.The lip segmentation,which is one of the most important steps in lip recognition,has a very important position in the lip reading system.This paper has carried out research and practice on lip segmentation.Due to the complexity of the background and the face,it is difficult to detect the lip region directly in the image.Therefore,in order to accurately detect the lip region,this paper designs a two-stage rough detection method for the lip region.Firstly,the Adaboost algorithm base on Haar-like is used to detect the face.Then,on the basis of the face images,the lip region is roughly detected according to the structural features of the face.In order to segment the lip image more accurately and quickly,considering the lip color has higher saturation and chromaticity,this paper proposes a method based on CIE-LUV color space to segment the lip.Since the initial contour of the active contour model(ACM)determines the segmentation effect,an adaptive initial contour algorithm is proposed.In order to improve the defect that chan-vese model is not suitable for segmenting uneven gray-scale images,an improved chan-vese model based on the LBF model's idea is proposed.The improved model adds local energy terms,which can reduce the influence of gray unevenness,And it can converge faster.After comparison experiments,the proposed algorithm has better segmentation effect than threshold segmentation and chan-vese model.With the increase of the accuracy of the lip segmentation,it will have a great influence on the effect of lip reading system.Therefore,this paper also has significance in application.