Sar Image Segmentation Using HMT Models Based On Bandelet Domain

Synthetic Aperture Radar (SAR) has the ability to take images in all the day and all the weather conditions, and SAR systems achieve good azimuth resolution, so it has been widely used in both military and civilian field.Image segmentation is a key technique for image processing. Its result will has significant effect on the sequential processing tasks such as classification and object recognition. Wavelet is the optimal base for the functions with point shape singularity, and its coefficients are sparse. But in high dimensional cases, wavelet analysis cannot take advantage of the geometrical features that data contained inherently. It is not the optimal or the sparsest representation of the function and always damages information of anisotropic edges in images.Wavelet-domain Hidden Markov Tree (WD-HMT) models are powerful tools for modeling the statistical properties of wavelet transforms. By characterizing the joint statistics of the wavelet coeficients, HMTs efficiently capture the characteristics of a large class of real-world signals and images. the wavelet domain hidden Markov model provides a natural framework for exploiting the statistical property of edges and ridges inherent in the real world images, especially grey-scale texture images, which are well characterised by their singularity structures. Within this framework, the histogram distribution of wavelet coefficients in the same scale is modelled as a two component Gaussian mixture model, and the persistence property of wavelet coefficients across scales is formulated as a Markov chain. So far, WD-HMT has been applied for many areas, such as image de-noising, image segmentation, and texture analysis and synthesis.Bandelet is a new multiscale geometric analysis tool for image representation. The remarkable aspect of Bandelet is the introduction of the geometry flow. As the coefficient distribution of Bandelet has the similar non-Gaussian distribution characteristic of high peak and heavy-tailed with the wavelet coefficient ,we introduce a kind of new method based on Banedlet domain HMT SAR segmentation, a finer to coarser HMT segmentation, which combine with the classification results of the three subbands for the 2D wavelet transform. The processes of SAR image segmentation is studied, and validated by experimental results.