High performance coding using a model-based bit allocation with EBCOT

The bit allocation procedure of JPEG 2000 is based on a rate-distortion curve directly computed from quantized and encoded wavelet coefficients. Thus, JPEG 2000 exploits at best the efficiency of its bit plane context-based arithmetic coder. However, encoding data which will not be saved in the final bit stream introduces complexity and JPEG 2000 bit allocation requires a lot of tests. We propose a new compression scheme using a low complexity model-based bit allocation followed by scalar quantizers with optimized deadzone sizes and the EBCOT bit plane coder to entropy encode the quantized subbands. The resulting compression scheme provides the same performances as JPEG 2000 with less complexity and simpler hardware implementation.