Invertible deinterlacing with sampling-density preservation: theory and design

摘要

A novel class of deinterlacing for intra-frame/field-based motion-picture coding, such as Motion JPEG2000, as well as an inter-frame-based coding without the support of interlaced scanning video, such as MPEG-1, is developed. This technique has two features: sampling-density preservation and invertibility. These features mean that the amount of deinterlaced pictures is not increased, and the original pictures can be perfectly reconstructed. This deinterlacing technique is a kind of sampling-lattice alteration and is regarded as a generalization of conventional field interleaving and field separation. With the help of multidimensional (M-D) multirate theory, it is shown that the design problem of such a system, that is, invertible deinterlacer with sampling-density preservation, can be replaced to finding a 2/spl times/2 multivariable polynomial matrix with a monomial determinant. This problem resembles the design of two-channel M-D maximally decimated perfect-reconstruction finite-impulse-response (FIR) filterbanks. The inverse system, which is referred to as a reinterlacer , is given as FIR when the dual deinterlacing system is FIR. A practical design procedure is provided by suggesting three constraints considered to be preferable: normalization, regularity, and vertical symmetry. The significance of the procedure is verified by showing some design examples of deinterlacing and reinterlacing filters. Simulation results show that the developed method causes fewer comb-shaped artifacts than conventional field interleaving.