Water-Bottom Video Coding Based on Coding- Oriented Reference Frame Generation

The video contents containing water scenes, such as seashore, water-sports, river-side, waving pool etc. are rich in sensation of presence. From the video coding point of view, those sequences are quite difficult to encode efficiently because of random movement and nonlinear deformation of the objects seen through waving water. This paper aims to tackle this problem by generating one additional frame from input video sequence. We additionally encode the frame and store as longterm reference frame and then encode the input sequence while referring to the additional frame. The key idea of it is to generate as good reference frame as possible in terms of motion compensation accuracy, and hence, in terms of coding efficiency. Our novel approach yielded BD-Rate reduction of 12 to 42% compared to VTM1.1 and 13 to 48% compared to VTM4.0, next generation video coding test model.