Researchers from the lab in Beijing have developed a system that can take your still photographs and automatically convert them into motion clips.
“A single photograph contains extremely rich content,” said Xian-Sheng Hua, the lead researcher on the project. “We wanted to use this content to build entertaining videos from still photos.”
In the Photo2Video project, the researchers used three key steps to automatically convert photos into video, taking their cues from traditional filmmaking techniques. The steps are story generation, framing scheme, and video composition.
Before completing these steps, the system selects a set of photos to include in the video, by browsing your archive of photos and clustering the images. It identifies photos that belong together, such as photos from the same event or trip, by analyzing semantic features.
The program uses face detection to identify similar faces, whether the photos were taken indoors or outdoors, in the city or a landscape such as mountains, the sea, or a meadow. It also determines which photos belong together by extracting meta data, such as the timestamp, from the photograph.