A Video Stitching System of Underwater Image

Abstract

Stitching underwater videos captured by a hand-held camera can essentially enhance the entertainment experience. However, underwater video is difficult to stitch due to its shortcomings in definition, visibility, shakiness and parallax. In this paper, we propose a video stitching system for underwater videos captured by mobile cameras. Our method consists of three components, underwater image enhancement, unified video stitching and 3D projection. The underwater image enhancement is a component based on Relative Global Histogram Stretching (RGHS). After the processing of this component, videos with higher definition and better visibility can be obtained, which will benefit the subsequent video stitching. The unified video stitching is a component based on background identification. Through background identification, we divide video stitching into two steps: background video stitching and foreground video stitching. This makes the stitched panoramic video more stable and the picture more natural, which is the key to heavy shakiness and large parallax problem. In addition, we have also studied the playback of the panoramic video to make it more immersive. We test the proposed system on videos that are captured by hand-held cameras when diving, and use SSIM, UCIQE, UIQM, and stitching score to quantitatively evaluate the generated panoramic video. The experimental results prove the effectiveness and robustness of our method in various cases, and the better video stitching results than existing methods.