Edge Detection of Noisy Images Combining Modulus Maxima with Improved Scale Product of Non-subsampled Shearlet Transform

In order to extract clearer and more continuous edges from a noise-contaminated image, and further improve the effect of edge detection, a method of edge detection is proposed based on modulus maxima and improved scale product of non-subsampled shearlet transform. Firstly, the multi-scale and multi-direction decomposition of noise-contaminated image is performed through non-subsampled shearlet transform (NSST) to get the high-frequency coefficients in NSST domain. Then the high-frequency coefficients in two adjacent larger scales are selected for the improved multi-scale product operations, and the NSST modulus maxima processing is carried out to get the binary image of edges. Finally, the isolated points are removed according to the regional connectivity and the accurate edge image is obtained. The simulative experimental results show that, compared with four edge detection methods (wavelet modulus maxima method, wavelet scale product method, non-subsampled contourlet transform (NSCT) modulus maxima and scale product method, NSST modulus maxima method), the method proposed in this paper has stronger ability of resisting noise, and the influence of texture is avoided effectively. The detected edges are clear, complete and with high continuity.