W e explore a n image compression scheme that does not rely o n the usual additive decomposition of a n input image into a set of predefined spanning functions. The proposed scheme only encodes implicit properties of the image and reconstructs a n estimate by constrained highpass energy minimization. This has particular advantages for very low bit rate perceptually oriented image compression. W e show experimental results in a context where we mainly target the smooth areas and the edges o f t h e image.
Yuk-Wong Choi, Nguyen T. Thao