Research Of Nonsubsampled Contourlet Transform And Its Application On Image Processing

Contourlet Transform is a new "true" two-dimensional representation for images, with the characters of multiresolution, directionality and localization. It has a wide application in image processing. But Contourlet transform is not shift-invariant, so it does not perform perfectly in some image processing. Nonsubsampled Contourlet Transform (NSCT) overcomes the shortage of Contourlet Transform, so it outperforms Contourlet Transform in image denoising, digital watermarking etc. NSCT is proposed in 2005, its theory is not perfect, and it is not yet widely used. We will research on the theory of NSCT and its application on SAR image denoising and digital watermarking. The main work and achievements of this thesis are summarized as follows:1. The basic theory and algorithm of Contourlet Transform and NSCT are introduced, and we make a comparison of this two transforms.2. We propose a NSCT based despeckling method for Synthetic Aperture Radar (SAR) images, derive in theory that the coefficients of log-transformed SAR images decomposed by NSCT obey general gaussian distribution, so adopt Bayesian shrinkage factor to estimate noise-free NSCT coefficients. Simulation and Experiments demonstrate that the visual quality of the results is excellent than other despeckling methods, in terms of both background smoothing, preservation of edge sharpness and textures. It also can avoid artificial impairments often introduced by critically subsampled transform.3. We propose a NSCT based adaptive digital watermarking method. This method establish a model of the human visual system (HVS) in NSCT domain, using visual masking effect of HVS to adaptive embed watermarking sequences. Simulation and Experiments demonstrate that the digital watermark not only the transparency can be guaranteed, but also has good robustness to noise additon, JPEG compression, cropping, filtering and image twisting etc. Furthermore, the watermark can be detected without original image. This is a perspective watermarking method.