Video Composition using Geodesic Distance

彬彬

1 Introduction

Video composition is an indispensable technique in the production

of many types of videos; nevertheless, it remains challenging problem

if the source video is normally captured without a blue or green

screen, especially when the object to be pasted between videos does

not have a clear boundary (e.g. water, gas, and fire). We will call

such a region a secondary foreground, and we propose a simple

composition method in which foreground, background, and secondary

foreground weights are determined using a geodesic distance

tramsform [Criminisi et al. 2010]. This method will run in

real time, but is nevertheless able to produce convincing results with

difficult secondary foregrounds.

2 Method

Figure 1(b) shows one of the user-defined trimaps for the source

video, which is the usual starting-point for image and video matting

processes. Let the binary masks Mf and Mb be one for foreground

and background region, respectively. We can compute the

geodesic distance transform [Criminisi et al. 2010] for each region:

Df (x) = D(x;Mf ;rI);Db(x) = D(x;Mb;rI): Figure 1(c)

shows the resulting geodesic distances Df and Db in blue and red

respectively. Intuitively, Df (x) is the distance from a point x the

boundary of the foreground, and Db(x) is the distance from x to

the boundary of the background.

The key feature of our approach is the additional secondary foreground

region. Let Pf (x), Pb(x), and Ps(x) respectively be the

probabilities that a pixel x belongs to the foreground, background,

or the secondary foreground. We can express these probabilities as

follows:

菜刀吻电线

提醒猪猪，千万不能让你看见

菜刀吻电线

“再次路过……”我造一个-----特别路过

菜刀吻电线

很有心,部分已收录自用,谢谢

C.R.CAN

既来之，则看之！

菜刀吻电线

好`我顶``顶顶

C.R.CAN

跑着去顶朋友滴铁

铁锹 3D打印店铺

3D技术领衔行业应用_技术创新促进产业升级


3D全息投影让乐器也疯狂




2012国际3D魔幻艺术中国特别展杭州揭幕





我国牵头制定智能电视和3D电视国际标准





美国4D影院模拟大风爆炸等效果



什么比Kinect精确100倍？_且看体感控制系统


Unity引擎滑稽3D跑酷玩儿命跑


非同一般的新GPS系统增强现实

tc

我看看就走，你们聊！

晃晃

很经典，很实用，学习了!