On the trade-off between contour-adaptive texture coding and lossy shape coding

Shape-adaptive texture coding is often being perceived as a means to achieve increased coding efficiency, e.g. due to a better representation of correlation properties along the boundary between two objects which should lead to a higher transform gain. However, if data rates demand a lossy encoding of contours, an improvement in transform gain becomes questionable. This paper investigates the potential coding gain which may result from shape-adaptive texture coding using DCT basis functions under the constraint of both lossless and lossy shape coding. For this purpose, a 1D texture model is derived which reflects synthetical as well as "natural" borders between objects.