Simple algorithm for correction of geometrically warped underwater images

摘要

A new algorithm for the turbulence degraded frames of an underwater object or scene to produce images with superior quality and higher resolution is presented. This approach is based on computing the centroid image of the sequence through an iterative scheme. A robust and fast optical flow technique is employed for pixel registration of all the frames against a reference frame and the pixel shift maps are determined. Using these shift maps, each distorted frame is restored through an image dewarping technique. A performance comparison between the proposed restoration approach and state-of-the-art is presented in terms of restoration time and accuracy. The effectiveness of the approach is demonstrated on both synthetic and real-world underwater videos.