Saliency Detection of Underwater Target Based on Spatial Probability

Due to the complex underwater optical environment, it is difficult for the classical detection algorithm to detect underwater targets accurately. A new underwater saliency detection model, inspired by the visual attention system of the human, is presented. First, the image is segmented into super pixels using the SLIC (simple linear iterative clustering) method and color region contrast feature is extracted based on them. Then, the target geometric center is located using Harris corner detection operator, it is used to describe the target space distribution feature in the form of center probability and all of the features are fused by the way of adaptive target location. Last, an optimizing saliency method is reported to highlight the foreground and weaken the background based on target space distribution feature, in addition, it encourages continuous saliency value using space smoothness. Experimental results show that the method reduces the influence of underwater noise and complex background on target detection effectively, and improves the validity and accuracy of target detection in underwater environment.