Data and rate adaptive quantization for joint image denoising and compression
Neelesh Kumar Gupta, E.I. Plotkin, M.N.S. Swamy · 2004
The techniques proposed for joint denoising and compression of images corrupted with additive white Gaussian noise are mostly based on Rissanen's minimum description length principle and tend to operate at a particular point (or a set of points) on the rate-distortion curve. These offer some compression along with denoising, but not a practical encoding solution. This paper suggests a simple adaptation of the zero-zone and the reconstruction levels of the uniform threshold quantizer based on the noise level in the image and the required compression rate. Context-based classification is also described for the noisy coefficients, and this raises the performance of the subband coder significantly.