Mahmoud R. El-Sakka and Mohamed S. Kamel, "An Edge-Preserving Neural Network For Image Compression", World Congress On Neural Networks, (WCNN'1994), pp. 59-64, May 1994, San Diego, California, USA.
Abstract
When conventional BP ANN is employed for image encoding the decoded images usually exhibit some degradation of the edges. This is due to the fact that edge pixels usually represent a small portion of the entire image and BP learning algorithms do not differentiate between edge and non-edge pixels. In this paper, a novel Edge-Preserving ANN learning algorithm is proposed. This learning algorithm pays more attention to edge information. The error between the computed and desired output value is multiplied by a weighting factor which is proportional to the amount of edge information in the corresponding input pixel. The algorithm is implemented and its performance is assessed by comparing it to the conventional BP.