Research On Lip Segmentation Based On Active Contour Model

In the modern society with computer science and technology are increasingly developing,human has more and more demands for scientific and technological life,Human Computer Interaction technology is also widely used in various fields.The emergence and application of these speech recognition systems have enriched human's intellectual life and improved their quality of life.However,speech recognition systems have some disadvantages.It can be finded that when in a noisy environment,the desired effect cannot be achieved.Because the recognition of these systems only rely on voice,the recognition rate will decrease linearly when there is noise or other sound interference in the use environment.To this end,researchers began to look for solutions.According to the experience of human communication,some researchers have proposed that speech content can be identified through changes in the shape of lip during speech.This method compensates for loss of auditory channel information through visual channel information.Therefore,more and more researchers have begun to devote themselves to the study of lip visual language recognition and have made remarkable achievements.This paper mainly researches the process of lip segmentation in lip language recognition.Before segmenting the lip area accurately,its need to locate the approximate lip area in the input images.This paper adopts a relatively mature face recognition algorithm to identify the face area in the input images.Then,in the face area image,the lip area is located according to the distribution of the facial features and the proportional relationship between them,and the lip image is obtained and used as an input for accurate segmentation.In order to make the final segmentation results more accurate and reduce the influence of other factors on the algorithm,this paper preprocesses the image before segmentation.First of all,in order to minimize the effect of light unevenness on the segmentation result,this paper performs the brightness equalization before segmenting image.In this paper,the original image is an RGB image.For ease of calculation,this paper performs luminance equalization for each color channel of the image,and finally combines the results to form a new image after equalization.In addition,this paper use a combined color space composed of U component in CIE-LUV space and C2 and C3 component of Discrete Hartley Transform to perform color space conversion on the image after luminance equalization,in which U component can highlight the difference between the lip area and the skin area,the C2 and C3 components retain more detailed information about the lip.Finally,in the combined color space image,the position of the four key points is determined according to the pixel value.Then two initial contour models are established respectively through these key points based on lip state.For the closed lip,the diamond is selected as the initial contour.In addition,we chose two semi-ellipsoids as the initial contour of the open mouth.In order to improve the segmentation accuracy,the initial contour of the inner lip is proposed in this paper,which reduces the influence of the internal area of the mouth on the result.From the experiment,it can concludes that the method used in this paper can complete the exact segmentation of the lip and can obtain accurate results.For open mouth,the results were obtained by the method used in this article can exclude other areas of the mouth,including only the lip,to ensure the accuracy of the segmentation.