A shape-adaptive transform for object-based coding

The efficient representation of image texture on arbitrarily shaped regions of support is an important problem to be solved in region-based coding schemes. One solution for smooth signals is the use of a linear transform, in order to achieve decorrelation and energy concentration for the input signal. In this paper, a general formulation for a class of shape adaptive transforms, that includes some known shape-adaptive methods as particular cases, is proposed and discussed. Within this general formulation, a new shape-adaptive, rectangular transform-based algorithm is investigated. The performance evaluation includes both theoretical decorrelation and distortion vs. rate results.