Automatic Segmentation of High Speed Video Images of Vocal Folds

[摘要] An automatic method for segmenting glottis in high speed endoscopic video(HSV) images of vocal folds is proposed. The method is based on imagehistogram modeling. Three fundamental problems in automatic histogrambased processing of HSV images, which are automatic localization of vocalfolds, deformation of the intensity distribution by nonuniform illumination,and ambiguous segmentation when glottal gap is small, are addressed. Theproblems are solved by using novel masking, illumination, and reflectancemodeling methods. The overall algorithm has three stages: masking, illuminationmodeling, and segmentation. Firstly, a mask is determined based ontotal variation norm for the region of interest in HSV images. Secondly, aplanar illumination model is estimated from consecutive HSV images and reflectanceimage is obtained. Reflectance images of the masked HSV are usedto form a vertical slice image whose reflectance distribution is modeled by aGaussian mixture model (GMM). Finally, estimated GMM is used to isolatethe glottis from the background. Results show that proposed method providesabout 94% improvements with respect to manually segmented data incontrast to conventional method which uses Rayleigh intensity distributionin extracting the glottal areas.

[学科分类] 应用数学