With uploaded to YouTube each minute, finding the right video without downloading it first is difficult. And video searching on mobile devices with slow internet connections can be frustrating. Help could be at hand in the form of a new preview interface that allows users to skim through a video before downloading it.
The search function on sites like YouTube often offers the viewer only a few frames from each video, making selecting the right result difficult, says Ben Falchuk at in Piscataway, New Jersey. 鈥淚t鈥檚 an extremely poor man鈥檚 preview,鈥� he says.
With Telcordia鈥檚 prototype, however, online or mobile users see many frames from a video, wrapped onto an animation of a 3D sphere that can be spun round at will to browse through the content.
Advertisement
The approach is something like the scene selection options on DVD menus 鈥� but Telcordia鈥檚 system allows the viewer to 鈥渮oom鈥� into a segment by tapping on a frame of interest, says Falchuk. 鈥淭he frames on the sphere are then reselected to correspond to the sub-scene of the video that the user is examining,鈥� he says.
With more and more people viewing videos on their phones, one of the advantages of this is that the interface allows the viewer to browse the video while downloading only about 3 per cent of the file. 鈥淚t saves the user a great deal of time,鈥� says Falchuk.
Keyframe success
The key to the software鈥檚 success will lie in the choice of sampled video frames. The simplest approach is to sample frames from the video at regular time intervals, says Falchuk, but there鈥檚 no guarantee that frames selected like this will give a good sense of what each segment contains. To ensure that only frames of interest appear on the preview sphere, other tricks must be used 鈥� for example, automatically choosing the first frame of a new shot as a sample frame.
One promising option the firm hopes to explore is crowd-sourcing the interesting bits of a video. By analysing the way that viewers watch a video 鈥� identifying which areas are frequently skipped, and which are rewound and watched again 鈥� Falchuk says it will be possible to produce an aggregated map of the 鈥渉otspots鈥� in the video.
, a digital multimedia researcher at the German Research Center for Artificial Intelligence in Kaiserslautern, thinks that advertisers may also see the benefit in the technology 鈥� it could be used to help ensure their ads match the contents of the videos, he says.
But there are still challenges in getting this sort of technology to work, says Ulges. To work effectively, the software must be able to automatically analyse each video posted online and produce a viewing sphere for it so that the sphere is available whenever a user鈥檚 search brings up that video. 鈥淭he question is whether the software can scale up to cope with the ever-increasing amount of [online video] content,鈥� he says.
The work will be showcased by the IEEE Communications Society at the (CES) in Las Vegas, Nevada, next month.
