Statistical modeling and threshold selection of wavelet coefficients in lossy image coder

An algorithm for using wavelet domain data quantization to improve the compression efficiency is presented. A conditional probability model of adjacent (in scale-spatial sense) magnitudes was applied as a better approximation of the wavelet coefficient dependencies than the marginal data distributions. This model was utilised in the threshold data selection and is proposed as a more effective uniform quantization modification than increasing the dead-zone. The same conditional model was used in quantization and encoding of the quantized magnitudes. Additionally, to fit the adaptive threshold value to local image features, estimation of significance expectation was included in the thresholding procedure. As a result, a more effective low-cost quantization scheme was constructed. It allows a significantly increase in image compression efficiency. An experimental rate-distortion curve shows the same distortion for decreased bit rates even up to 20% in comparison to standard uniform quantization