Abstract: In the video detection of parameters such as ship size, the synchronous movement of the stern ripples seriously affects the accuracy. A novel algorithm of ships saliency detection for inhibiting the stern ripples was proposed based on video sequence. Firstly, the algorithm utilized a histogram-based contrast (HC) method to define HC saliency map for inland waterway by using color statistics of the input image. Then, it performed super-pixel segmentation on original image to get several sub-regions and used regional spatial relationship to improve HC saliency test results, which was named as regional saliency map. Finally, by the initialization of GrabCut algorithm with the regional saliency map, the iterative process was added by erosion and dilation operations to get close to the target edge, so that the moving ship was extracted. Experimental results showed that the proposed approach could effectively restrain the stern ripple, accurately detect the ship in inland waterway, and its accuracy was up to 94.6%.

Key words: stern ripples, saliency, super-pixel segmentation, GrabCut algorithm