Abstract:

Focused on the issue of ghosting and artifacts due to moving foreground in video stitching, a video stitching method for minimizing parallax artifacts is proposed. The proposed method firstly improves the energy function according to the foreground region, grayscale difference, texture and color difference of the overlapping region image, and uses the energy minimization dynamic programming technique to find the optimal suture of the non-moving foreground region. Secondly, the Visual Background Extraction（VIBE） is used to detect the moving object and the selected optimal suture is updated in the subsequent video image. Finally, the fusion region is adaptively adjusted according to the color saturation difference, and the adaptive fusion method is used to realize the video image. The theoretical analysis and simulation results show that the method can effectively suppress the parallax artifacts caused by the moving foreground and there is no obvious seam.

Key words: video stitching, energy function, optimal stitching, background extraction, adaptive fusion

摘要：

针对视频拼接中由于移动前景引起的重影和伪影问题，提出了一种视差伪影最小化的视频拼接方法。该方法根据重叠区域图像的前景区域、灰度差异、纹理和颜色差异对能量函数进行改进，同时采用能量最小化的动态规划技术找到非移动前景区域的最佳缝合线；利用视觉背景提取（VIBE）检测移动物体并在随后的视频图像中更新最佳缝合线；根据色彩饱和度差异自适应地调整融合区域，采用自适应融合方法实现视频图像的融合。实验结果表明，该方法可以有效地抑制由移动前景引起的视差伪影且其拼接图像无明显接缝。

关键词: 视频拼接, 能量函数, 最佳缝合线, 背景提取, 自适应融合