视频压缩中的运动估计
我有一个与上面标题相关的问题。根据我的阅读，该技术的理论是采用先前的参考帧并基于该参考帧对当前帧进行估计。然后，逐块地在参考帧中搜索与当前帧中的预测的匹配。然后您将获得参考帧和当前帧之间的运动矢量。但这对我来说并没有什么意义，因为最后你仍然只能得到真实参考帧（我说是真的，因为我假设它是第一个空间帧内编码参考帧）和预测帧之间的向量。如果我对技术的理解错误，有人可以向我解释一下吗？
谢谢
I have a question related to the title above. From what ive read, the theory of this techinique is to take a previous reference frame and make an estimation of the current frame based on the reference frame. Then, on a block by block basis, search for a match in the reference frame with regard to the prediction in the current frame. Then you get a motion vector between the reference frame and the current frame. This doesnt really make sense to me though because at the end of it all you are still only getting a vector between the true reference frame (i say true because i'l assume its the first spatially intra coded reference frame) and a predicted frame. Can someone please explain to me if im understanding the technique wrong??
Thanks
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（1）
运动估计结果通常不仅仅是“参考帧和当前帧之间的运动矢量”。这部分被称为“全局运动补偿”，并且确实在帧级别有一些专用算法。然而
，运动估计的思想是不仅如此。就是在图像中找到一个移动的元素，这意味着即使“框架”不移动，你仍然想检测到框架中从左到右移动的汽车。
这部分是“逐块”完成的，称为“块运动补偿”。
因此，运动估计结果是一系列向量，为小像素块组提供（对于 MPEG，通常为 16x16，甚至对于 H264，甚至为 4x4（尽管最后一个的布局确实很复杂））。
Motion estimation result is typically more than just a "motion vector between reference frame and current frame". This part is called "global motion compensation", and indeed has some dedicated algorithms for it at frame level. http://en.wikipedia.org/wiki/Global_motion_compensation
However, the idea of motion estimation is more than that. It is to find a moving element into an image, which means that even if the "frame" doesn't move, you still want to detect a car into the frame which moves from left to right.
This part is done "block by block", and is called "Block motion compensation".
Hence, the motion estimation result is a serie of vectors, provided for groups a small pixels blocks (typically 16x16 for MPEG, or even 4x4 for H264 (although the layout is really complex for this last one)).