WMFLICM: A Robust Algorithm for SAR Image Segmentation Using Hybrid Spatial Information

Abstract

ABSTRACT Synthetic aperture RADAR (SAR) imaging is an important means for performing the task of remote sensing. Image segmentation is one of the most crucial steps done prior to the classification and identification of different regions and objects present in the acquired images. The presence of multiplicative speckle noise in SAR images makes the tasks of image processing extremely challenging. The fuzzy C-means (FCM) segmentation technique and its variants perform satisfactorily for images corrupted by additive noise, but these algorithms and other intensity-based conventional methods do not show encouraging results in case of speckle-contaminated SAR images. Segmentation performance for SAR images can be improved by incorporating spatial context information. Also, heavily contaminated SAR images cannot be effectively processed based on the intensity feature alone. Image textural features prove to be robust and more appropriate for the purpose of segmentation of SAR images. Hence a hybrid methodology, Weighted Membership Fuzzy Local Information C-Means (WMFLICM), is proposed in this research work, which uses spatial context information by incorporating both implicit and explicit neighbourhood information in terms of feature similarity in local window. In the proposed method, wavelet energy based feature is used to represent the textural information for the images generated by SAR sensors. Experiments conducted on synthetic as well as real SAR images demonstrate that the proposed algorithm with enhanced spatial information is more effective than other methods used for segmentation of SAR images.