Seafloor visual saliency evaluation for navigation with BoW and DBSCAN

In this paper, we try to combine Bag-of-Words (BoW) with Density-Based Spatial Clustering of Applications with Noise (DBSCAN) together for one kind of sparse representation in the seafloor visual saliency evaluation. Properties in the water, due to the large amount of acoustic noises, sonar signals are easily polluted and interfered during image collection, and the sonar images usually diverge from the true underwater environment or degrade the accuracy of the measure, so sparse representation has been taken into consideration for underwater sonar image visual saliency evaluation. Our method is a simple and computationally efficient extension of an orderless BoW and DBSCAN model, and it shows consistently and significantly improved performance on sparse representation in the seafloor visual saliency evaluation.