Underwater salient object detection jointly using improved spectral residual and Fuzzy c-Means

In this paper, we propose an underwater object detection method, which uses improved spectral residual (SR) saliency detection and fuzzy segmentation. We adopt a two-phase mechanism, which divides visual object detection into detecting saliency map and image segmentation to obtain "proto object". We compare the logarithmic spectrum differences between optical images in the atmosphere and in the water. Combining with the absorption characteristics of the propagation of light in water, we use the logarithmic spectrum of underwater images and logarithmic spectrums in R, G and B channels to generate new logarithmic spectrum, so as to highlight more object information and obtain better saliency map. Then, using Fuzzy c-Means (FCM) clustering method to segment saliency map, we gather better similar information of the object and highlight the entire body of the objects. We tested the effectiveness of our method in underwater object detection in different underwater optical environments. The results show that our method can eliminate most of the background noise and improve the accuracy of underwater visual object detection.