Transform image coding with edge compensation
Tetsuo Shimono, Yoshio Kanda, Hideo Kitajima, Yoshihiko Ogawa · Electronics and Communications in Japan (Part I Communications) · 1991
Abstract A method to prevent edge degradation caused by orthogonal transform image coding is proposed. The method contains three processes: 1) obtain a differential image between the original and low‐pass one; 2) extract significant pixels concerning the edge degradation; and 3) add these pixels to compensate for the edge degradation. Since an edge has a spatial length, five or more connected pixels whose value exceeds a threshold are selected in the differential image as significant error pixels. Furthermore, using human visual perception that is very sensitive to the existence of an edge but very loose in its exact value and its exact location, the additional information (overhead) for compensation is saved. The significant edge information is quantized three levels corresponding to positive error, negative error, and no error. Then edges are classified in four directions after slight geometric approximations. Four one‐dimensional images with three‐level value are assigned by Huffman code. Run‐length coding is used jointly for the zero error pixels. The additional information for the edge compensation requires a few bits, but the image is reconstructed very clearly.