Towards video laryngostroboscopy-based automated screening for laryngeal disorders

This paper is concerned with kernel-based techniques for automated categorization of laryngeal colour image sequences obtained by video laryngostroboscopy. Features used to characterize a laryngeal image are given by the kernel principal components computed using the N -vector of the 3-D colour histogram. The least squares support vector machine (LS-SVM) is designed for categorizing an image sequence (video) into the healthy, cancerous and noncancerous classes. The kernel function employed by the LS-SVM is defined over a pair of matrices, rather than over a pair of vectors. The classification accuracy of over 85% was obtained when testing the developed tools on data recorded during routine laryngeal videostroboscopy.