Font Size: a A A

Research On Lip Segmentation And Inner And Outer Contour Extraction

Posted on:2018-11-30Degree:MasterType:Thesis
Country:ChinaCandidate:J W FuFull Text:PDF
GTID:2428330596989255Subject:Electronic and communication engineering
Abstract/Summary:PDF Full Text Request
With the rapid development of computer science,pattern recognition and artificial intelligence,the application of human-computer interaction based on the visual speech system has aroused the interest of many researchers in recent years.As fundamental and key technologies,lip segmentation and lip contour extraction have become more and more important.The realization of lip-reading system is separated into two steps,lip segmentation technology and lip contour extraction technology.Thus,in this paper,these two steps are presented respectively in different chapters.The lip segmentation is mainly based upon the fuzzy clustering framework.In this paper,the major contributions of the proposed approach are 1)Two ellipses are modeled to fit the region of the lip and inner mouth according to the segment result solely depend on the color information.The spatial information aids to differentiate pixels with similar color and located in different region(including outside mouth,inside mouth and in the lip region);2)A competitive mechanism is introduced in the objective function and optimization procedure to remove those nonexistent inner mouth components and thus the proposed approach can obtain accurate segmentation results for lip images with both open and closed mouth.In comparison with these existing methods,the proposed algorithm obtains more accurate segment result.Particularly for those images with open mouth,the inner mouth components do less interfere in the accuracy of segmentation.For the lip contour extraction technology,because of the clear segment result of the inner mouth components with the aid of the competitive mechanism,not only the outer lip contour,but the inner lip contour can also be extracted while there are few approaches can extract inner lip contour accurately.Based upon the membership map derived from the segment result,according to the Maximum Likelihood Estimation(MLE),with the gradient ascent optimization procedure,the inner and outer lip contour can be extracted successfully with geometric constraints by quadratic curves.Experiments show that the approach provides accurate extraction results.
Keywords/Search Tags:lip segmentation, spatial penalty term, competitive learning, inner and outer contour extraction
PDF Full Text Request
Related items