MRF Segmentation Of SAR Sea Ice Image Combined With Speckle Reduction And Region Growing

Sea ice is an important part of the global climate system and sea ice information has a significant meaning for the monitoring of climate change and human activities. Synthetic Aperture Radar is an effective tool of sea ice monitoring and can all-day, all-weather access to a wide range of high resolution image. SAR has been widely applied in the study of sea ice.The segmentation methods based on MRF model are frequently used in SAR sea ice image segmentation. However, due to SAR sea ice images are seriously affected by speckle noise, the MRF model can not accurately describe the statistical characteristics of the image so that leading to the decrease of the segmentation accuracy. Hence, this paper proposes a new sea ice segmentation algorithm based on region-level MRF model combining with a speckle reduction region growing model (SRRG-MRF).1. Establishing the Speckle reduction region growing (SRRG) model. The SRRG model includes two parts:constructing images speckle reduction regional representation and region growing based on the gray similarity of adjacent regions. The images speckle reduction regional representation achieved by the speckle reduction bilateral filtering (SRBF) algorithm and watershed transform. In the case of images affected by the speckle noise seriously, SRRG model can effectively inhibit the over-segmentation and position target edge accurately.2. Combining region-level MRF model with the SRRG model to obtain effective SRRG-MRF segmentation algorithm. The SRRG model combined with the region-level MRF can significantly reduce the optimization search space and prevent the energy function optimal solution in a local minimum value, so as to obtain the accurate segmentation results.3. The proposed SRRG model and the SRRG-MRF segmentation algorithm has been evaluated using several synthetic SAR sea ice images corrupted with varying levels of speckle noise as well as the Liaodong Bay sea ice image acquired by the RADARSAT-2 and the Labrador coast C-band sea ice image obtained by the SIR-C.The results show that the SRRG-MRF algorithm can effectively improve the segmentation accuracy compared with the existing region-level MRF segmentation algorithm and demonstrate the effectiveness and feasibility of the proposed method.