Converting video to high-resolution photo

Contact person: Dmitriy S. Vatolin (dmitriy@graphics.cs.msu.ru)

MSU Graphics & Media Lab (Video Group)

This filter is intended to create high quality still images from video.
The improvement is achieved by using information from neighbouring frames. The resulting images have also higher resolution than original video.

The main problem of this filter is slow speed, but the algorithm has very good parallelization potential.

Examples

Here is the comparison of MSU Video2Photo filter results with original frames increased by factor of 2 and bicubic resize method.
Only frozen frames and corresponding frames of source videos are shown, other frames are omitted.

First example is the 4th frame of test sequence "news":


Source frame

Bicubic-resized frame Processed with MSU Video2Photo frame


Source frame

Bicubic-resized frame Processed with MSU Video2Photo frame

Further pictures are fragments of frame from the "mobile" sequence:


Source frame

Bicubic-resized frame Processed with MSU Video2Photo frame

Another example is the 4-th frame of "foreman" sequence:


Source frame

Bicubic-resized frame Processed with MSU Video2Photo frame

And a frame from the "src_ref_625" sequence:

Source frame Bicubic-resized frame Processed with MSU Video2Photo frame

See also

Detailed description of MSU Video2Photo filter

Team