This thesis investigates two topics: (1) Fast search block motion estimation algorithms for digital video encoding; and (2) Motion detection/estimation based deinterlacing for digital video interoperability....[ Read more ]

This thesis investigates two topics: (1) Fast search block motion estimation algorithms for digital video encoding; and (2) Motion detection/estimation based deinterlacing for digital video interoperability.

Fast search for block motion estimation is modeled as a checking points (the candidate motion vectors) allocation problem that is subject to the motion vector statistics. A statistical framework is established to define, derive and evaluate the fast search block matching algorithms. Numerical measures are designed to study the short and long term properties of Motion Vector Distribution (MVD) for natural video, since MVD is used as a priori knowledge for the Checking Point Pattern (CPP) optimization. Three classes of algorithms are proposed: (1) Adaptive CPP algorithms that adapt to the frame-based MVD. (2) Fixed CPP algorithms that are suitable for hardware implementation. and (3) Dynamic CPP algorithms, where the CPP adapts to both long term and local MVD. Simulation results of the proposed algorithms are compared to the results of the full search and previous algorithms.

Based on the analysis of the residual energy in the frequency domain, a phase-correction filter is proposed to improve the motion compensation for interlaced signals. Conversely, reliable motion detection/estimation can be achieved using the phase-corrected field to replace the field with a different parity for motion estimation. Filter design trade-off is investigated with regard to filter tap, hardware cost and residual energy suppression. Hardware cost-effective deinterlacing algorithms are proposed based on motion detection/estimation. Experimental results are compared to the results of previous algorithms. Furthermore, an arbitrary ratio resizer is proposed for digital video interoperability.