Perceptual Relevance Measure for Generic Shape Coding

Abstract

We address a fundamental problem in subjective reconstruction quality by introducing a perceptual relevance measure (ADM) for generic vertex-based shape coding. Different from traditional absolute distance measure (ADM), our proposed measure systemically considers the turn angel and two adjacent segments for corresponding vertex visual significance calculation. We embed our proposed measure into top-down and bottom-up frameworks, stages of both vertex selection and adjustments, and class one and class two distortion measure definitions. The experiment results show that our proposed measure can significantly improve subjective reconstruction quality as well as objective rate-distortion performance, especially for object shape with sharp salience.