Image stitching method based on content optimisation in Virtual Reality
虚拟现实中基于内容优化的图像拼接方法

摘要

图像拼接是虚拟现实技术中虚拟场景构建的核心技术之一，而通过图像拼接得到的全景图像中总是存在伪影、重影、畸变等问题。为了快速鲁棒地拼接图像并得到高质量的全景图，提出了一种基于内容优化的图像拼接方法。通过局部内容相似性保持和线性相似性保持的优化规则可以保证拼接前后图像中的基本内容以及线段的不变性，从而最大限度减少全景图像中伪影、重影、畸变的产生。该方法是基于粗匹配+精细匹配的框架，首先根据匹配特征点获取图像单应变换模型实现图像初步对齐，然后再根据优化规则进行内容优化以减少伪影和畸变的产生得到高质量的全景图像。通过和相似方法的实验对比结果表明，提出的方法匹配精度高，拼接图像产生的伪影畸变少，能快速获取高质量拼接图像。

Abstract

Image stitching is one of the core technologies for virtual scene construction in virtual reality technology, and there are always artifacts, ghosting, distortion and other problems in the panoramas obtained by image stitching. In order to quickly and robustly stitch images and obtain high-quality panoramas, an image stitching method based on content optimisation is proposed. The optimisation rules of local content similarity preservation and linear similarity preservation can ensure the invariance of the basic content as well as the line segments in the image before and after stitching, so as to minimise the generation of artifacts, ghosting, and distortion in the panoramia. The method is based on the framework of coarse matching + fine matching, which firstly obtains the image homography transform model according to the matching feature points to achieve the initial alignment of the image, and then performs the content optimisation according to the optimisation rules to reduce the generation of artifacts and distortion to obtain a high-quality panorama. The experimental comparison results with similar methods show that the proposed method can quickly obtain high-quality spliced images with less artefacts and distortions.