Sar Image Change Detection Based On Mean Shift Pre-Classification And Fuzzy C-Means

Ronghua Shang,Kaize Xie,Michael Aggrey Okoth,Licheng Jiao
DOI: https://doi.org/10.1109/IGARSS.2019.8898464
2019-01-01
Abstract:In order to reduce the influence of noise and obtain better change detection effect, this paper proposes a method for SAR image change detection based on mean shift pre-classification and fuzzy C-means. First, the original image is pre-classified based on mean shift clustering. As a clustering method with non-parametric density estimation, mean shift can effectively maintain the edge information of the object, and can smooth the pixel intensity of the same type of object to reduce the influence of noise on change detection. Then, the difference map is generated by the log-ratio operator and classified into changed area, uncertain area, and unchanged area. After the adjustment, the pre-classification is performed by mean shift and the difference map is generated. Finally, the improved FCM algorithm is used to classify the difference map to generate change detection result map. The effectiveness of the proposed method is verified by experiments with different contrast algorithms on real SAR image datasets.
What problem does this paper attempt to address?