DCT computation based on variable complexity fast approximations

We investigate input dependent, variable complexity algorithms to compute the DCT. The basic goal of these algorithms is to not compute those DCT coefficients that will be quantized to zero. These algorithms exploit the fact that for compression applications (i) most of the energy is concentrated in a few DCT coefficients and (ii) as the quantization step size increases an increased number of coefficients is set to zero and reduced precision computation of the DCT may be tolerable. Thus we propose two classes of algorithms, the first one selectively prunes the DCT computation while the second uses an approximate computation, without floating point multiplications, that is matched to the quantization level selected.