Design of 2D-multiple notch filter and its application in reducing blocking artifact from DCT coded image

The reconstructed image from highly compressed JPEG data has noticeable degradation due to blocking artifacts. Linear, space-invariant filtering is inadequate to remove these artifacts. So, space-variant or adaptive filtering is required. These schemes are based on edge information extracted from the received blocky image, which contain false edges due to blocking artifacts. Based on analysis in frequency domain, it is found that artifacts in decoded image have periodic structure. These periodic-textures become more obvious after gradient operation since discontinuities due to block coding are highlighted in gradient image. According to signal processing theory, periodic texture will result in some peaks in frequency domain. Here, peaks are due to false edges in gradient image or due to blocking discontinuities in decoded image. A two-dimensional (2D) multiple-notch filter can kill these peaks to reduce the artificial edges due to blocking artifacts in gradient image. Here, a 2D IIR multiple-notch filter is designed from one-dimensional multiple bandpass filter using simple algebraic method. This approach is an extension of method used by Pei at el. (1994) in the domain of 2D. Discontinuities due to blocking artifacts are more in monotone area so this filter can be applied directly. Thus, improvement in the performance of space-variant/adaptive filter by removing of false edges from edge of blocky decoded image and reduction of blocking artifacts directly in monotone area, are the new possible application of multiple-notch filter which is considered in this work. Computational complexity is low because filtering is performed in DCT domain.