Enhanced Pixel-Based Video Frame Interpolation Algorithms

Abstract

In this paper, we compare three motion compensated interpolation (MCI) algorithms: adjacent-frame motion compensated interpolation (AFI), wide-span motion compensated interpolation (WS-TH), and wide-span motion compensated interpolation with spatial hinting (WS-TH+SH). The latter represents an extension to WS-TH by adding spatial hinting to the generation of motion vectors. The methods are quantitatively compared with the objective of optimizing interpolated frame quality relative to control interpolated frames. This is important because for high-resolution large flat-panel displays, frame transition coherence becomes a critical factor in assessing the quality of the user's viewing experience. To enhance MCI, the encoder should attempt to exploit long-term statistical dependencies, precisely estimate motion by modeling the motion vector field, and superimpose efficient prediction/interpolation algorithms. Computer simulations using artificially generated video sequences demonstrate the consistent advantage of both WS- TH and WS-TH+SH over AFI under increasingly complex source scenes and chaotic occlusion conditions.