Image compression based on low-pass wavelet transform and multi-scale edge compensation. Part II: evidence and experiments

Summary form only given. The scalability and recognition ability of the edge detection of MSEC can be observed, in which two kinds of edges are handled separately, and, for each kind of edge, the detector responds to the exact scale only. For example, the apparent large-scale edges in the original image are kept undetected until an appropriate decomposition level is reached. Compensated images are considerably smoothed since fine details are systematically removed. Such smoothing is seriously different from the traditional smoothing method in both the idea inside and the effect outside. The low-pass wavelet transform works since the compensated image contains much less high-frequency energy. By the low-pass wavelet transform, the system is able to reach and process edges at the next larger scale. So, the encoder outputs multi-scale primal sketches, which are coded in a modeled way instead of pixel by pixel. The encoder also outputs the final smooth background, which can be readily coded by traditional methods. The decoder synthesizes the image according to the received information on the multi-scale primal sketch and the background.