Postprocessing of decoded color images by adaptive linear filtering

Linear filtering of image subbands for low complexity postprocessing of decoded color images

“…In [1], image adaptive linear minimum mean squared error (LMMSE) filtering was proposed as an enhancement layer color image coding technique that exploited the statistical dependencies among the luminance/chrominance or Karhunen Loeve Transform (KLT) coordinate planes of a lossy compressed color image to enhance the red, blue, green (RGB) color coordinate planes of that image. In the current work, we propose the independent design and application of LMMSE filters on the subbands of a color image as a low complexity solution.…”

“…As an alternative to joint coding of the transform coordinate planes, [1] proposed to enhance the coordinates of a pixel by applying a LMMSE filter with support extending across all coordinate planes as well as over the spatial neighbors of the pixel. This LMMSE filtering based enhancement layer coding technique was demonstrated to successfully exploit the statistical dependencies among the transform coordinate planes and thereby efficiently reduce the quantization noise.…”

“…67]) used to transform the decoded luminance/chrominance (or KLT) coordinate planes to get reconstructed RGB color coordinate planes only covers the filtered pixel, and does not cover the sites of its neighbors. The work in [1] showed that inclusion of the correlated neighbors of the filtered pixel in the filter's support is essential for the enhancement layer coding gain obtained by the LMMSE filter.…”

“…The number of elements in the correlation matrix grows with the square of the size of the filter support. Secondly, since the original image data needed for the LMMSE filter design is available at the encoder, but not at the decoder, the method proposed in [1] requires the coding and transmission of each LMMSE filter coefficient to the decoder as overhead. Even though the marginal return (ratio of the decrease in distortion to the increase in rate) at a given rate due to the application of the method is significantly larger than that yielded by any color image coding method, highest marginal returns are obtained for small filter supports which cover only the most correlated neighbors.…”

“…Since, as pointed out in [1], the correlations between the coordinates of the filtered pixel and the coordinates of the neighboring pixel is an important determinant of the LMMSE filtering gain, we include in the filter support of each subband only those decoded luminance/chrominance coordinates of neighboring subband coefficients that are sufficiently correlated with the corresponding decoded coordinates of the filtered subband coefficient.…”