ABSTRACT

In natural images, there usually exist many textures, lines, and edges. Along the edges, sample intensities change smoothly; but across the edges they will change dramatically even with a small shift from the edges. This kind of correlation shows a strong anisotropic feature because it depends upon not only the distance of samples but also their link orientation. Furthermore, the link orientation often goes in all directions rather than only the vertical or the horizontal. Traditional transforms such as the discrete cosine transform (DCT) and discrete wavelet transform (DWT) cannot handle this directional correlation well and energy compaction is not fully achieved. Therefore, many methods have been proposed to exploit the directionality in images/intra frames in terms of transforms and predictions.