A Speckle Filtering Method of SAR Images Based on Wavelet Analysis

The images of synthetic aperture radar(SAR) have been widely used in many important fields.Like any coherent imaging system,SAR systems are subject to speckle effects.This noise is conspicuous in single-look SAR images which disturb image interpretation and target classification.Therefore,to extent the use of SAR images,effective speckle reduction techniques are highly desirable. Based on the principle of multiresolution wavelet analysis,a new filter for suppression speckle in SAR is proposed.For images,an algorithm similar to the one-dimensional case is possible for two-dimensional wavelets and scaling functions obtained from one-dimensional ones by tensorial product.This kind of discrete two-dimensional wavelet transform leads to a decomposition of approximation coefficients at level j+1 in three components:the approximation at level j and the details in three orientations (horizontal,vertical,and diagonal).The "approximate image" at level j is composed of the low frequency parts in both row and column directions of approximate image at level j+1.And the there "detail images" contains its high frequency components separately in horizontal,vertical,diagonal directions.The approximate image is iteratively decomposed into four sub-images level by level.A pyramidal wavelet decomposing structure is thus constructed.It is tes-tified that the subspaces level of 4 or 5 is adequate for multiresolution wavelet analysis of an image (1024×1024).At each stage of the multiresolution pyramid,the approximate signal is low-pass filtered and decimated.And the wavelet coefficients are computed by high-pass filtering.It is proved that the signal energy concentrates on the wavelet coefficients of higher absolute value,and that the coefficients of high absolute value is more valuable than that of low absolute value in the signal wavelet reconstruction.It is also demonstrated that the effect of speckle is more serious in higher frequency parts,and the speckle decrease sharply as the image is low-filtered.So it can be considered that the speckle in 4 or 5 level approximate image can be ignored.The detail images of each subspace in which the ratio of signal to speckle is comparatively low are processed.The speckle of detail image at high frequency level is suppressed comparatively more by soft-threshold technique while the speckle at low frequency is suppressed comparatively less.The filter's threshold is computed according to the number of level,the variance and the length of each detail images. The result image which is reconstructed by the processed detail images of each level is estimated through the effect of vision,the quantitative analysis such as mean,standard and variance,and the analysis of preservation of edge and detail.The filtering results of ERS-1 SAR image and image with simulated speckle show that the method proposed in this paper is satisfying in visual appearance,speckle suppression and detail preservation.