SAR image despeckling by iterative non-local low-rank constraint

Abstract

SAR image despeckling is a fundamental problem which degrades the performances of SAR image automated analysis and information extraction. In this paper, we proposed an SAR image despeckling algorithm by iterative non-local low-rank constraint. Dual weights are used to make the best use of the non-local similarity relation. Logarithmic transform is firstly applied to gain independent variance noise. Then under the non-local methodology, low-rank constraint is realized by weighted nuclear norm minimization which has lower complexity. For accurate non-local information, weighted averaging is utilized defined by the index of the similarity. Compared with the sate-of-art despeckling algorithms, our method has preferable performance and lower computational complexity. Both the subjective sense and objective indicator show the capacity of this method on noise reduction and detail preservations.