We are proposing "reconstructive video coding" framework with super-resolution※1 technology for prospective ultra high-definition videos. The resolution of the images is reduced adaptively depending on the transmission bandwidth and the picture content before the images are encoded by a conventional encoder such as AVC/H.264. The decoded images are reconstructed to the original resolution by up-sampling them with super-resolution techniques.
- Picture-adaptive image reduction
- This technique adaptively reduces the resolution of input images taking account of the transmission bandwidth and the picture content. The successive conventional video encoder further compresses the reduced images preserving sufficient image qualities.
- Image reconstruction using super-resolution
- The super-resolution reconstructor up-samples the decoded reduced images into the original resolution, which utilizes the inter-frame motion and the spectral aliasing※2. We have developed super-resolution techniques that work properly in this video coding framework.
- Reduction/reconstruction control for video quality optimization
- The optimization of the reduction/reconstruction parameters are also performed on the reduction stage. The reconstruction stage can precisely and stably restore the images with the parameters provided as side information.
In the works
We will further improve the image quality by refining image reduction and super-resolution techniques, and we intend to use this framework for encoding Super Hi-Vision videos.
- ※1Super-resolution:Technology for increasing image resolution
- ※2Aliasing:False patterns and jaggy edges that appear when the image has finer textures than the pixel intervals
Configuration of reconstructive video coding