Visual Maritime Attention Using Multiple Low-Level Features and Naïve Bayes Classification

This paper presents a framework for Visual Attention Detection in maritime scenes. The focus is to provide an early processing stage for high resolution images captured by maritime surveillance platforms. The framework groups multiple low-level features that are designed specifically for maritime scenarios with different distance measurements. Integrated in the framework is a detector for sea and sky that aids in background segmentation. A Naive Bayes Classifier is used to produce Attention Maps of the input images. Experiments using ground truthed images show the technique is effective on a large dataset of maritime images and outperforms state of the art generic saliency detectors.