Analysis of motion compensated filters concerning motion correctness and accuracy
作者:
Highlights:
•
摘要
This paper analyses linear filters designed for motion compensated interpolation or prediction. So far, the design of motion compensated filters was based on sampled moving scenes or sequences of single motion compensated images. In this paper, such filters are derived from the three-dimensional model of motion compensated sequences. Assuming pure translational movement in the scene as a commonly used simplification of the real scene movement, motion compensated sequences are described in the spatio-temporal and three-dimensional frequency domain. The differences to stationary sequences and typical characteristics dependent on the correctness and the accuracy of the assumed motion are pointed out. On this basis, motion compensated filters are optimized for a range of velocities in the first method and for definite velocities in the second. Both methods are discussed in relation to filter performance and filter length and are compared to those presented in the literature. The impact of subpel accurate motion on the filter characteristic is treated in detail. As an example, motion compensated filters are designed for up-conversion of interlaced sequences. The theoretical results are confirmed by simulations.
论文关键词:Motion compensation,Motion compensated sequence,Motion compensated filter,Subpel accuracy,Prediction,Interpolation,Deinterlacing,Generalized sampling theorem
论文评审过程:Received 20 April 1996, Available online 30 June 1999.
论文官网地址:https://doi.org/10.1016/S0923-5965(98)00039-3