Elastic registration for videostroboscopic images of the larynx

摘要

Videostroboscopy is an examination method during which a video-recording of the vocal folds can be obtained. This examination is very important because it yields a permanent record of the moving vocal folds and it allows the diagnosis of abnormalities which contribute to voice disorders. In this paper a new algorithm based on simulated annealing (SA) is used to register/match the videostroboscopic images of the larynx. This algorithm operates on the contours representing the vocal folds. The matching process is done through minimizing a cost function, which consists of two parts. The first captures the requirement that the distance between two pixels that are matched should be minimized, while the second part requires the smoothness of the displacement vector field. Four parameters are used to characterize the SA algorithm: 1) the acceptance ratio $chi@, which is set to give a high initial temperature $Tau$-0$/ to start with, 2) a small positive number $delta which controls the decrement of the temperature, 3) the length L of the Markov chains that brings the system to equilibrium at each temperature, and 4) a stopping parameter $epsilon@S that defines the final state/configuration of the system. The performance of this matching algorithm is demonstrated on simulated and real videostroboscopic images. It is shown that this algorithm is successful in matching videostroboscopic images when the deformation is severe.