Efficient Frame Interpolation

Real-time frame interpolation for increasing the frame rates of videos

The refresh rates of today’s mobile phones and computers are getting higher and higher, but the refresh rates of many videos are still stuck at less than 30FPS. Video interpolation algorithms can effectively bridge the gap between the video frame rate and the refresh rate of the display device.

My target is to design a real-time frame interpolation method which can increase the frame rates significantly while maintains the details of the frames.

My method is based on the flow-based frame interpolation algorithms. The whole process can be divided into two part, flow estimation and fusion. With two frames as input, the optical flows from the inexisted middle frame to the two reference frames are estimated respectively. Then the reference frames are warped based on the optical flow. The final frame is fused based on the warped images. Shown below is an interpolated example.

Original Video Frames.
8x Interpolation & Smoother Motion.

My method achieve 35.36 of PSNR, and is much more faster than all the other methods.

My Method is Faster when Interpolate Good Frames.

My master degree thesis is based on this project.