Abstract: The noisy image is first decomposed into multi-level wavelet to obtain the corresponding coefficients, and a filter is constructed by correlating the multi-level wavelet coefficients. Then, the filter is applied to an original level with smoothed wavelet coefficients to get noise-free image. Finally, the filtered image is thresholded to produce the edge image. Experiment result shows that this approach can get rid of much more noise than conventional edge detection methods.