Feature matching for underwater image via superpixel tracking

Feature matching is fundamental to many vision tasks. Due to the low visibility of images in underwater environments, traditional pixels-based matching methods suffer from miss-matching or error-matching. Recently, Superpixel based features have been applied to image feature analysis. However, most of existing methods dedicate to rectified stereo matching with images captured in the air. This paper presents a novel feature matching scheme aiming at underwater images. It targets the un-rectified image pair from the video sequence. The Superpixel matching process is fulfilled with multiclass labelling based on Markov Random Field (MRF). Experiments show that the proposed method produces competitive performance.