A color segmentation scheme for object-based video coding

Abstract

A color segmentation scheme is presented for use in an object-based video coding environment. The proposed technique employs the HSV (hue, saturation, value) color space to partition a video-phone type sequence into a facial area and a set of arbitrarily shaped regions. The facial region is identified by utilizing the a priori knowledge of the skin-tone distributions along with a series of fuzzy shape-based membership functions. The remaining regions are segmented using a recursive 1D histogram thresholding procedure. The technique is robust, computationally inexpensive, and allows for a higher coding efficiency. Furthermore, it may be used to support the content-based functionalities addressed in future object-based coders.