Unsupervised Change Detection in Remote sensing Image Based on Image Fusion in Nonsubsampled Shearlet Transform Domain and fuzzy k-means clustering

Abstract

In order to improve the detection precision and shorten the detection time, a novel unsupervised change detection method based on image fusion in nonsubsampled shearlet transform(NSST)domain and fuzzy k-means clustering is proposed in this paper. Frost filter is used to reduce the noise of the experimental images. The proposed neighborhood ratio operator and the common log-ratio operator are used to obtain difference images. In order to utilize fully the complementary information of the neighborhood ratio and the ratio images to obtain a better difference image, a novel fusion strategy in NSST domain is proposed. Since there are still noise in the difference images, the image denoising method with adaptive Bayes threshold in the NSST domain is applied to the high frequency coefficients of the difference images to reduce the noise. And then the proposed fusion strategy is applied to the low frequency bands and the denoised high frequency bands for getting the fused difference image. The change detection map is obtained by clustering the fused difference images utilizing k-means algorithm into two disjoint classes: changed and unchanged. The experimental results clearly show that the proposed detection operator has better detection performance and shorter running time, compared with the other reported algorithms.