Lossless, near-lossless and lossy adaptive coding based on the lossless DCT

This report proposes a new coding method, which realizes lossless, near-lossless and lossy coding with a unified single algorithm. The method is based on a lifting-structured lossless DCT, which maps an "integer" input vector to also an "integer" output vector. The proposed method can be used as not only a high performance "lossless" encoder but also a "lossy" encoder compatible with conventional DCT-based methods such as JPEG and MPEG. In addition, it has a "near-lossless" mode in which compression ratio is better than lossless mode retaining quite good quality of decoded images. In "lossless" mode, only the lossless DCT and an entropy coder are used. In "lossy" mode, transformed coefficients are quantized and then entropy coded in the same manner as the conventional method for compatibility. In "near-lossless" mode, signals are quantized before DCT to avoid SNR degradation due to "rounding" operations in the lossless DCT.