The Research On The Speckle Reduction Algorithm For Sar Images

The contents of this dissertation focus on the research for the algorithm of the speckle reduction for the Synthetic Aperture Radar (SAR) images. With its ability to image the earth's surface in nearly all weather conditions, together with its high spatial resolution, the SAR has been playing more and more important role at fields of geosciences, bionomics and hydrology, etc. However the SAR, which is a kind of coherent system, can not avoid producing a random pattern, named speckle noise in its images. Moreover, unlike the additive white Gaussian noise, the speckle noise is multiplicative in nature, which degrades the merit of the SAR images and affects their further application seriously. To overcome this serious drawback in the SAR images, it is essential to reduce speckle before the next procedures. So the speckle reduction becomes one of the key steps in the SAR image processing. This dissertation makes deep research on how to improve the performance of the despeckling algorithm.Firstly, we study the ideas and structure of the adaptive speckle filter which is based on the local statistic characteristics in the space domain and propose a modified Frost speckle filter based on the anisotropic diffusion (ADFF). Unlike the conventional Frost filter, the weighting coefficients of the ADFF are anisotropic or edge sensitive, which are adjusted according to the information of the edge in the local processing window. The anisotropic characteristics help the ADFF improve the performances in terms of the speckle reduction and get the better edges preservation and enhancement.Secondly, we make the wavelet transform for the SAR image and analysis the statistic characteristics of the wavelet coefficients. First, a wavelet-based algorithm named the H-G Wavelet-based algorithm is proposed. In this algorithm, a new model which combines the Hidden Markov Tree (HMT) model and the Gaussian Markov Random Fields (GMRF) model is used to depict the characteristics of the wavelet coefficients. With this model, the algorithm is able to capture the correlations of the inter-scale, inner-scale and inter-direction for the wavelet coefficients, therefore it can get the better despeckling performance. Second, we present a simpler despeckling algorithm named the Cauchy MAP wavelet-based algorithm. In this algorithm, the Cauchy prior is applied to model the distribution of the wavelet coefficients for the log-transformed reflectance of SAR images. In addition, the introduction of the"second kind statistics"makes the parameter estimation become more efficient and it also improves the performance and computer efficiency for the algorithm.Finally, follow the ideas of the image multiscale geometric analysis, we make the contourlet transform for the SAR image and study the statistic characteristics of the contourlet coefficients. The HMT model and the GMRF model are extended from the wavelet transformed domain to the contourlet one and a new algorithm named the CHMT-G Contourlet-based algorithm is proposed. On the other hand, a new parameter named Inter-Direction Variation Coefficient (IDVC) is introduced to reflect the correlation of the inter-direction of the contourlet coefficients. With the adoption of the new model and the IDVC parameter, the correlations between the contourlet coefficient and its generalized neighborhoods are captured completely in this algorithm. The experimental results show that the algorithm has the better performance especially on the preservation of the details and the visual effects of the despeckling image.To verify our proposed new algorithms, the simulated speckled images and the real SAR images are used in the experiments. We evaluate the performances of our algorithms with both the filtering visual results and some quantitative evaluation indices.