SAR Image Change Detection Based on Mathematical Morphology and the K-Means Clustering Algorithm

Abstract

Synthetic aperture radar (SAR) images have been applied in disaster monitoring and environmental monitoring. With the objective of reducing the effect of noise on SAR image change detection, this paper presents an approach based on mathematical morphology filtering and K-means clustering for SAR image change detection. First, the multiplicative noise in two SAR images is transformed into additive noise by a logarithmic transformation. Second, the two multitemporal SAR images are denoised by morphological filtering. Third, the mean ratio operator and subtraction operator are used to obtain two difference images. Median filtering is applied to the difference image based on a simple combination of the two difference images. Since an accurate statistical model for the difference image cannot be easily established, the results of change detection are clustered using the K-means algorithm. A comparison of the experimental approach with other algorithms shows that the proposed algorithm can decrease the detection time and improve the detection result.