Segmentation based image coding with l-infinity norm error control

In this paper we present a segmentation based technique for lossy image coding with minmax error control. The technique is an improvement of a scheme presented in \cite{DL_linfty_04}. In the base scheme, the image is recursively and adaptively divided into rectangular regions with a binary tree decomposition. Every region is approximated by mean of a bilinear surface, whose values on the four corners are encoded. It is well known that binary tree decompositions can generate many adjacent regions that contain the same texture, but have been separated higher up in the decomposition of the hierarchy. Inspired by the work of Sukla et al. (\cite{ImprovQuad}), we here investigate the possibility of jointly coding such similar adjacent regions, so as to reduce the bit rate.