The Despeckling Of SAR Image Based On The Curvelet Transform

With the development of the Synthetic Aperture Radar(SAR), it's widely used in many domains, however, the SAR images contain speckle because of the coherence effects between the backscattered signals when imaging, the speckle noise can typically be modeled as multiplicative. The speckle reduces the image quality and the spatial resolution of SAR image. In order to understand and interpret the SAR image, it's necessary to despeckle but keep the details at the same time. The complexity of the scatter targets make the SAR image doesn't have an accurate model, so the traditional denoising methods don't work effectively when despeckle. However, the appearance of multiscale geometric analysis provides a new way to despeckle effectively.In this paper, one of the multiscale geometric analysis tools, curvelet is used to despeckle the SAR image. Curvelet is a multiscale transform with frame elements indexed by scale, location and orientation parameters, the frames can represent edges and other singularities along curves efficiently. Based on this, three new despeckling methods are proposed, the main innovative points are as follows:1) The statistical characteristics of curvelet coefficients of the SAR image is analyzed first, then a mixture density of two zero-mean Gaussian functions is proposed to fit the actual histogram of the coefficients of every subband. Based on this, a Bayesian shrink factor is derived to shrink the curvelet coefficient, lastly, the mean filter and the nonlinear anisotropic diffusion are used to deal with the reconstructed image to overcome the shortcomings of curvelet.2) Contraposing the aggregation property of the curvelet coefficients in every subband, a shrink factor is proposed. The stationary wavelet is introduced to deal with the point targets because the curvelet doesn't work well on it, in the wavelet domain, the pertinence of the wavelet coefficients is used, so the point targets on the edges are well maintained, at last, the shrink of the difference image between the original image and the result image of wavelet is used to get the point targets.3) A continuous threshold function is proposed to improve the hard-thresholding function proposed by Stark in the curvelet domain. In the curvelet domain, the coefficients in the finest subband are dealt with using different principles twice, and reconstruct the image twice so as to keep the weak texture well, and the complex diffusion is used to compensate the mean value. Experimental results demonstrate that the three methods can despeckle the SAR image effectively, and keep the details well at the same time.