A novel procedure for land masking in ocean-land segmentation from SAR images

摘要

Land masking is one essential preprocessing technique for ship recognition from synthetic aperture radar (SAR) images. This paper presents a novel land masking procedure based on wavelet transform(WT). Low-pass (L-P) images for several wavelet scales are produced. Two wavelet correlation criteria are developed to fuse useful information at multiple scales. The first criterion makes land segmentation at each scale via an adaptive threshold approach, followed by a local correlation process to select land pixels as many as possible while discard false alarm pixels; The second criterion is to fuse L-P images via a modulus correlator for speckle suppression, contrast enhancement and better region homogeneity, so that the following segmentation can be made easily. The final results are obtained by morphological processing and area filling. Experimental results from real SAR image processing demonstrate the usefulness of the proposed methods.