图像融合是视频拼图中获得高质量合成图像的重要步骤。利用梯度信息引导插值融合重合区域较小的两幅图像已经取得较好的效果,然而对于视频拼图而言,相邻两帧图像之间重合区域大,且多帧图像之间交叠重合,给这种方法的实际应用带来限制。为解决这一问题,设计梯度域内同时融合多帧图像的策略,利用图像的自然边界将拼图平面进行区域划分,在各区域内分别定义引导矢量场和边界条件,实现各划分区域的一次性融合。为克服拼图平面的过划分问题,提出相应的帧选择方法和区域合并算法。将该策略用于视频拼图实验,获得了视觉效果良好的合成图像,并提高了融合效率。
Image blending is a crucial process to achieve quality composite picture making in video mosaic. The method of guided interpolation by a guidance vector works well for blending two images with a relative small overlapping region in image mosaic. However, it is hard to apply the method to video mosaic, due to the large overlapping region shared by every two consecutive frames and the complex overlapping conditions among multiple frames. To deal with this problem, a framework to synchronously blend multiple images in the gradientdomain is proposed in this article: first, the mosaic plane is partitioned into several regions by the boundaries of each image; then, the guidance vector and boundary conditions are defined for each region; finally, the blending process is carried out for every region synchronously. To avoid over partition of the mosaic plane, a frame selecting scheme and a region combining algorithm are proposed. Applying the proposed framework to video mosaic achieves a fine composite picture with less consumptive computation.