Modelling quantization error from quantized Laplacian distributions

Abstract

Post-processing, restoration or transcoding of JPEG and MPEG compressed images require a quantization error model of discrete cosine transform coefficients. A method is presented to recover the original probability density function (pdf) parameter from quantized data, assuming that the original pdf of the coefficients is Laplacian. Quantization errors are calculated from the Laplacian model.