Adaptive Alpha-Trimmed Correlation based underwater image stitching

Abstract

Image stitching (Mosaicing) is a technique of combining multiple underwater images to generate a composite image with a much larger field of view. With its advancement in various imaging applications, underwater image stitching, in particular presents its own challenges. Underwater images suffer from clouding effects and grayed out colors as a result of absorption and scattering of light. Such complexities hinder the image stitching process causing misalignments and non-uniform illumination. The paper presents a system that combines multiple underwater images to generate a fused image, overcoming the impediments of a visible seamline. The presented method is based on a novel stitching technique called Alpha-Trimmed Correlation. The system is designed to adapt to the difficulties encountered in underwater image stitching. The performance of the stitching algorithm is tested against its ground truth, both subjectively and quantitatively. The quantitative analysis is performed using non-reference underwater quality measures such as UICM, logAMEE, and CTSUIQM.