Kee-Koo Kwon

Blocking artifacts reduction algorithm in block boundary area using neural network

A novel algorithm is proposed to reduce the blocking artifacts of block-based coded images by using block classification and feedforward neural network filters in the spatial domain. In the proposed algorithm, we classify the block boundaries into four classes based on a statistical characteristic of neighborhood block. For each class, adaptive inter-block filtering performs in block boundaries. That is, for the smooth region with blocking artifacts, we use the two-layer neural network with error backpropagation algorithm, and for the complex region with blocking artifacts, we use the linear interpolation method. Experimental results show that the proposed algorithm gives better results than the conventional algorithms from both the subjective and objective viewpoint.

Blocking artifacts reduction in block-coded images using self-similarity

Transform coding is the most popular image compression method. However, for high compression it produces noticeable blocking artifacts in the decoded image. This paper presents new algorithms for reducing blocking artifacts. The algorithms use piecewise self-similarity within different parts of the image as a priori to give reasonable modification to the block boundary pixels. The proposed postprocessing algorithm includes block classification in transform coefficient domain, and each block boundary area smoothing is performed by searching similar blocks. In experiments it outperforms an existing method and gives significant subjective and objective improvement over the decoded image.

Postprocessing algorithm in block-based coded images using wavelet transform and adaptive MLP

A novel postprocessing algorithm for reducing the blocking artifacts in block-based coded images is proposed using an adaptive neural network filter (NNF) in the wavelet transform domain. We use two characteristics, knowing the positions of the blocking artifacts in each band and discriminating between the horizontal and the vertical blocking artifacts using the wavelet transformed signals. At first, after performing a 3-level wavelet transform of the decompressed image, a different one-dimensional (1D) or 2D NNF is used to reduce the blocking artifacts according to the characteristic of each level. Experimental results show that the proposed algorithm produces better results than those of conventional algorithms, both subjectively and objectively.