Research and improving on speckle MMSE filter based on adaptive windowing and structure detection

Abstract

The appearance of speckle makes it difficult for image segmentation, classification and target detection. By analyzing MMSE filter and the enhanced one, it can be seen that although these filters are better trade-off between noise reduction and fine detail preserving, a basic premise is required for all statistic filters that the sample mean and variance of a pixel is equal to its local mean and variance based on pixels within a neighborhood surrounding it. It is validated only if the filtering area is large enough to assure the sample calculation robust and doesn't contain edge features, linear features or point targets for stationary assumption. An improved filter based on adaptive windowing and structure detection is proposed in this paper. This filter outputs the mean intensity of filtering area in homogeneous region. Maximum homogeneous filtering area decided by adaptive windowing technique is required to keep sample calculation robust. Structure feature such as line, edge and point target may appear in heterogeneous region and this invalidates the basic premise of stationary for all local statistic filters. Point target is preserved owing to special distribution and maximum window subset of linear and edge feature is decided by using different ratio detectors. Filtering process is accomplished calculating local statistics in window subset determined. Experiments on simulated image and real SAR one demonstrate that the improved MMSE filter proposed for speckle noise is superior both in noise reduction and in fine detail preserving.