基于格林坐标和改进最佳缝合线的图像拼接技术

Image stitching technology based on Green coordinates and improved seam estimation

为了应对大视差场景的图像拼接问题以及提高拼接图像的自然度,提出了一种基于对极几何约束和格林坐标的图像拼接算法。首先,用对极几何模型代替单应性矩阵作为随机抽样一致(RANSAC)算法的筛选模型,弥补单个全局单应性矩阵处理大视差图像时的不足,提高特征点匹配的精准度;然后,以一幅图像为参考,对另一幅图像进行基于格林坐标的预变形;其次,借助弹性鲁棒扭曲的思想消除预变形图像和参考图像之间的投影偏差;最后,将超像素分割的思想应用到缝合线的找取过程中,避免缝合线穿过物体重要部分,实现用最佳缝合线拼接图像。实验结果表明,所提算法能够有效地避免伪影的出现,能更加精准地实现图像对齐,使拼接后的图像更加自然。改进的算法使缝合线从超像素边界穿过,达到保持图像局部重要信息完整性的效果,可应用于电力传输与运维场景视觉监控领域。

In order to solve the problem of image stitching in large parallax scenes and improve the naturalness of stitched image,an image stitching algorithm based on epipolar geometric constraint and Green coordinates was proposed.Firstly,the epipolar geometric model was used instead of the homography matrix as the screening model of the RANdom SAmpling Consensus(RANSAC)algorithm to make up for the shortcomings of a single global homography matrix when processing large parallax images,and improve the accuracy of feature point matching.Secondly,taking one image as a reference,another image was pre-deformed based on Green coordinates.Then,the projection deviation between the pre-deformed image and the reference image was eliminated by the elastic robust distortion.Finally,the idea of super-pixel segmentation was applied to the stitching process of seam for avoiding the seam passing through the important part of the image,and the seam estimation was used to seamlessly stitch the image.The experimental results show that,the proposed algorithm can effectively avoid the appearance of artifacts,can achieve more accurate image alignment,and make the stitched image more natural.By using the improved algorithm,the seam passes through the super-pixel boundary to preserve the integrity of local important information in the image.The proposed algorithm can be applied to the scene visual monitoring in the field of power transmission,operation and maintenance.