Speckle Reducing Algorithms Based On SAR Images

Synthetic Aperture Radar has been widely used in military and civilian fields because it’s able to work from dayto night under all weather conditions.Due to SAR imaging reason,speckle exists in SAR images,which will hinder image interpretation,so it is of great significance to study how to reduce speckle.In this paper,three modified algorithms are proposed from three aspects:spatial filter,Speckle Reducing Anisotropic Diffusion and non-local means.The main works of this paper are as follows:(1)A modified Frost filter algorithm based on rectangle quad-windows is proposed.Based on an idea of rectangle bi-windows,a structure of rectangle quad-windows is proposed.By combining its edge strength maps and the weight of Frost filter,the modified algorithm can improve pixel estimate at edge areas via distributing different weights in local window.Experiments show that the modified algorithm not only reduces speckle but also preserves edges.(2)A modified SRAD algorithm based on the minimum-ration of neighborhood pixels is given.The weight of Frost filter is regarded as a diffusion coefficient,and we define tha a factor is combination of instantaneous coefficient of variation and speckle scale function.According to theoretical analysis,the modified diffusion coefficient can achieve isotropic diffusion in homogeneous regions.According to the minimum-ration of central pixel to neighborhood pixels in local window,the diffusion degree of all directions is guaranteed.The function of perpendicular to the gradient direction that enhances edges is obtained.Experiments demonstrate that combination property of the modified algorithm is better than those of other algorithms.(3)A modified non-local means algorithms based on ratio and space distances are provided.A ratio distance is modified and its validity and robustness are then proven.In the algorithm,combining modified ratio distance,the central pixel values and central pixel’s space location of image patches is proposed that form modified weight formula to estimate pixel values.Experiments illustrate that the modified algorithms have good performances in speckle reducing and edges-preserving.