This paper proposes an unsupervised approach for detecting regions of interest (ROI) based on the K multiplicative model (KMM) and superpixel segmentation in SAR imagery. KMM, with method of moments parameter estimation, mainly aims at detecting the darkcontain ROI on the sea surface. And the superpixelbased segmentation accomplishes edge extraction of ROI through clustering superpixels of ROI. The approach in this paper significantly outperforms adherence to boundary, achieving the error rate of less than 10% for airborne SAR imagery, and less than 2% for TerraSAR imagery. This approach also performs well in the computational efficiency, processing 104 pixels in 0.53 seconds on average. Experiments, accomplished by resorting to high resolution of the Lband airborne synthetic aperture radar (SAR) data and the TerraSARX data, demonstrate strong robustness and high effectiveness based on the proposed approach.