Image Denoising And Fusion Based On Non-Subsampled Contourlet Transform

Multiscale Geometric Analysis (MGA) is a tool of image sparse representation which developed on the basis of wavelet transform. It not only has the properties of multiresolution, local time-frequency property that inherited from wavelet transform, but also meets the features of anisotropy, low redundancy and robust. This thesis emphasizes the study on the representative Non-Subsampled Contourlet Transform (NSCT), and then applies its properties, such as translation invariance and geometric flow, to denoise and fuse the infrared and visible images.In image denoising, an algorithm based on NSCT is proposed. Firstly, the threshold average method is adopted to process the low frequency parts of NSCT coefficients. Then, the hierarchical adaptive hard threshold processing combine with the geometric flow properties of NSCT are used to deal with the high frequency parts of the coefficients. The simulation results show that the Peak Signal to Noise Ratio (PSNR) of the denoised image obtained by the algorithm has a0.5-4dB enhancement compared with the classical wavelets and contourlets denoising algorithms. Also the subjective visual effects are improved,In image fusion, several multi-scale image fusion methods and fusion rules have been analyzed and contrasted. An algorithm based on NSCT is proposed.We can complete the fusion of low frequency coefficients on the premise of suppressing the noise effectively by introducing the thresholds on the basis of traditional weighted average method. For the high frequency parts, the method that combined max-coefficient-choosing with local area rules is utilized and the image details are well protected.The experimental results show that the fusion algorithm can keep the contour information well and can inhibit the "virtual shadow".