Rare events detection algorithm: Application to sonar images

In this work, a new algorithm for rare events detection is proposed. Different from traditional accuracy oriented approaches [1], our approach employs the dressed sigmoid function presented in our previous work [2] to estimate the presence of rare events in the image. Our detection methodology is performed in two steps. In the first instance, a quick search for rare events using a sliding window is performed. At the second step, we propose to improve the efficiency of the detection method using the entropy criterion as we exploited it in our segmentation approach. Different results are tested on a set of SAS (Synthetic Aperture Sonar images and synthesized images.