The present work describes an advanced method for image preprocessing to improve the automatic detection of the glottal space from laryngeal images obtained either with high speed or with conventional video cameras attached to a laryngoscope. Images are filtered using an anisotropic diffusion technique that combines smoothing properties with image enhancement qualities. The preprocessing technique improves the performance of the previous system based in watershed transform and merging. Results show that 38% of the mismatches in delineating the glottis are fixed or reduced. 111 larynx images have been segmented to obtain the glottal area, 11 of the 29 previous errors have been corrected.

Index Terms. segmentation, preprocessing, anisotropic diffusion, glottis