Filtering level-set model based on saliency and gradient information for sonar image segmentation

The quality of the acquired sonar image is generally poor because of the influences of various marine environments, mainly in terms of low resolution and incomplete target edges. To address the difficulties of sonar image segmentation caused by the reasons above, this paper proposes a filtering level-set model for underwater sonar image based on saliency information and gradient information. The model consists of the following three modules. First, the filter fusion module can selectively smooth the noise of sonar image, which reduces the difficulty of subsequent segmentation. Next, the local enhancement module can effectively improve the over-segmentation problem caused by the weak boundary of the target in sonar image. Finally, the level-set segmentation module is proposed in this paper. Weakening the shadow on the results of the above modules can obtain the initial contour closer to the target boundary, which makes the evolution results more accurate. The experimental results show that our model is effective, accurate, and superior to the state of the art.