Extracting underwater object region proposal using BING method

In recent years, the emergence of underwater robots has enhanced our ability to understand and explore the underwater world. In order to inspect and maintain the underwater targets, the underwater robot must first check the presence of the target in the complex underwater scene. Due to the illumination attenuation and scattering in water, underwater images generally appear fuzzy and in low contrast, this problem makes the traditional color and texture features are not discriminative for underwater. To address this problem, propose to extract the binarized normed gradient (BING) feature for underwater objects, which is quite efficient. Then this robust feature is applied to train a classifier to find the underwater object proposal. Compared with the state-of-the-art algorithms of region proposal, our method takes only 0.3 seconds to process an image. In the experiments, it shows the application of BING for underwater object region proposal can reduce time complexity and improve the real-time performance of underwater target detection.