Algorithm of Motion Field Optimization for Temporal Frame Interpolation
Keywords:
True Motion Estimation, Optimization Problem, Temporal Frame Interpolation, Motion Field, True Motion ModelAbstract
Purpose: Development of a true motion estimation algorithm for temporal frame interpolation. Results: The paper proposes a new
true motion estimation algorithm which considers both the model of true motion of objects in video sequences and the cases when
this model is incorrect, e.g. at the boundaries of the objects. The algorithm is based on iterative search of the local minima for the
optimization problem introduced in the framework of the model. The search procedure assumes coarse-to-fine refinement of motion
from large objects to small ones, motion reliability estimation and reliability-based smoothing of the motion field. The paper evaluates
the algorithm for the problem of frame rate up-conversion on a set of standard test sequences. It is shown that a simple interpolation
algorithm based on the proposed motion estimation method demonstrates performance comparable to the existing frame rate upconversion schemes based on more sophisticated heuristics. Practical relevance: The proposed algorithm can enhance the visual quality of temporal frame interpolation (a key component in modern TV receivers) which is an important task for video processing systems.