Research On Key Image Processing Algorithms Bsed On Contourlet

Based on computational harmonic analysis, a novel Multiscale Geometric Analysis (MGA) theory was proposed to capture the geometrical structure in visual information efficiently, it can achieve optimal approximation behavior for2-D piecewise smooth function, thus obtaining highly efficient representation and processing methods. MGA tools includes Brushlet, Wedgelet, Beamlet, Ridgelet, Curvelet, Bandelet, Contourlet and Nonsubsampled Contourlet Transform (NSCT), etc. MGA theory and its applications still need further research and development. In this dissertation, the author mainly focuses on the research of Contourlet transform and the key techniques of its applications in image processing. As the latest MGA tool, contourlet and nonsubsampled contourlet are the "true" two dimensional sparse representations, nonsubsampled contourlet also has the "Shift Inviance" property, which provide much better anisotropy, multiresolution, directionality and localization properties for2-D signals than existing image representation methods. The main research work in the dissertation is as follows:1. For contourlet transform and its novel extension-nonsubsampled contourlet transform, we provide a detailed analysis to its basic principles, implementation schemes, advantages and disadvantages, which can be seen as the foundation of following image processing applications. The varieties of advantages of multiscale geometric analysis make it able to capture high dimensional singularities in natural image efficiently, and also consistent with the perception and integration principle of receptive field of visual cortex in human vision system. Therefore, they are more appropriate for various image processing tasks. It can be seen that by using contourlet or nonsubsmapled contourlet with proper scheme, better performance would be expected.2. A novel medical image fusion algorithm based on local statistics in Contourlet domain is proposed. All fusion operations are performed in Contourlet domain. Based on the tree structure among parent and children coefficients in Contourlet domain, the Contourlet contrast measurement is developed. It is proved to be more suitable for human vision system. Other fusion rules like local energy, weighted average and selection are combined with "region" idea for coefficient selection in the low-pass and high-pass subbands. The final fusion image is obtained by directly applying inverse Contourlet transform to the fused subbands. Extensive fusion experiments have been made on CT and MR medical images, both visual and quantitative analysis show that comparing with conventional image fusion algorithms, the proposed approach can provide a more satisfactory fusion outcome.3. A novel multifocus image fusion algorithm based on region statistics in contourlet domain was developed. According to the optical imaging principle, the defocused imaging system can be characterized as a lowpass filtering. For contourlet can handle intrinsic features in natural image much more effectively than wavelet. Original images was first decomposed by contourlet transform, fusion operation was implemented for all directional subbands in each scale, different region based fusion rules were used for lowpass and highpass subband, they are local energy and regional variance. Experimental results show that compared with traditional algorithms, the proposed multifocus image fusion algorithm can obtain a more satisfactory outcome.4. A geometrically robust image watermarking algorithm based on nonsubsmpaled contourlet transform (NSCT) and scale invariant feature transform (SIFT) is proposed in this paper. The SIFT features points are first computed from the host image, in which the suitable SIFT points for watermark embedding are selected and tested according to their magnitude changes after being modified. NSCT was adopted by virtual of its advantages over traditional wavelet transform and other multiresolution geometric analysis tools. NSCT decomposition is implemented on the SIFT feature circle region. The lowpass coefficients within the circle region are modified to embed the watermarking bits. To enhance watermarking detection performance, we proposed an equal area based circle pattern and improved odd-even quantization algorithm. Experimental results demonstrate that the proposed scheme is robust against the geometric attacks like rotation, scaling, cropping and moving, as well as the general processing operations.5. Based on the speckle noise modeling for underwater laser image, an adaptive speckle reduction algorithm using nonsubsampled contourlet transform (NSCT) is proposed. The statistical model for speckle noise is first analyzed to obtain a simple and tractable solution in a closed analytical form. Gaussian distribution for speckle noise and a general Gaussian distribution are adopted to model the statistics of contourlet coefficients in logarithmically transformed laser images. Then based on the maximizing the a posteriori estimation with the assumption that speckle noise is spatially correlated within a small window, we utilize a locally adaptive Bayesian processor whose parameters are obtained from the neighboring coefficients in highpass subbands. Experimental results show that comparing with classical wavelet method, the proposed algorithm shows a superior performance in suppressing the speckle noise and retaining geometrical structures of the image.6. Available image edge detection schemes based on the spacial domain or wavelet transform domain cancapture only limited directional edges in images. A novel multiscale edge detection algorithm based on nonsubsampled contourlet transform (NSCT) was proposed in chapter7. The input image is first decomposed into NSCT domain, different edge detection techniques are applied in lowpass subband and highpass subbands:a spatial domain algorithm like Canny operator is used in lowpass subband, every coefficient in t he subbands is thresholded by comparing against the adaptive thresholds. Based on the directions of each directional subband and its gradient, modulus maxima of the transform coefficients are obtained by comparing the modulus amplitudes of the samples and the neighborhood coefficients on the direction of the gradient. The scheme requires lower complexity without complicated link operation, while edge detecting performance is improved. The results show the proposed algorithm is superior to the image edge detection method based on wavelet modulus maxima by approximating better the edges of images and has lower computational complexity.