Harmonic postprocessing to conceal for transmission errors in DWT transmitted images

摘要

The discrete wavelet transform has become a dominant method in transform coding, but transmission of the resulting coefficients has lacked efficient postprocessing to recover from block loss in the lowest frequency wavelet subband. This paper relates the lost coefficients to the surrounding valid ones by the Laplace equation on a square grid. The properties of its solutions, viz. harmonic functions, ensure uniform and stable performance of the recovery scheme.The method is compared with interpolative functions already used for block loss recovery, such as the bicubic Coons surface (the only reported method to recover lowest frequency subband block loss) and NURBS, used in DCT domain block loss recovery. Experiments on standard test images show uniform superiority in both visual quality and PSNR measurement.The method can reconstruct both coefficients in the wavelet domain and pixels in the image domain. We compare results experimentally and analytically, to guide practical application.