An innovative, high quality and search window independent motion estimation algorithm and architecture for MPEG-2 encoding

A high quality adjustable complexity motion estimation algorithm for video encoders

“…The final motion vector points to the macroblock in the reference frame which is possibly the most similar to the macroblock under consideration in the current frame. This algorithm is 99 % less complex in computations than Full Search and its encoded picture quality has been found better than Full Search [10]. Even state-of-the-art research shows that SLIMPEG performs better than its competitors because of its sub-pixel accuracy [17].…”

“…However, the need of a mechanism of early termination and the usage of mixed templates in accordance with motion type makes these algorithms more complex. An implementation of "Substantially Light Motion Picture Estimator for mpeG" (SLIMPEG) has been described by Rovati et al in [10] for MPEG-2 and later by Alfonso et al in [3] for MPEG-4. Given that SLIMPEG [10] performs similar or better than optimal Full Search in terms of picture quality and it is being employed in industry [11], it has been used as the reference motion estimation algorithm in this paper.…”

“…An implementation of "Substantially Light Motion Picture Estimator for mpeG" (SLIMPEG) has been described by Rovati et al in [10] for MPEG-2 and later by Alfonso et al in [3] for MPEG-4. Given that SLIMPEG [10] performs similar or better than optimal Full Search in terms of picture quality and it is being employed in industry [11], it has been used as the reference motion estimation algorithm in this paper. We suggest two techniques of partially blind prediction for motion estimation for reducing complexity in motion estimation process.…”

Computational-complexity scalable motion estimation for mobile MPEG encoding

“…Up to now, the RME algorithms have been used on GOP structures with fixed M and B-frames were not considered for long-term tracking of the motion. A more sophisticated approach for ME is presented in [10], featuring a two-step estimation process and enhanced vectorfield prediction. The first step of this approach is a coarse RME to pre-estimate the forward vector-fields within an SGOP.…”

“…Besides special block-matching criteria such as [7], computational complexity scalability has been introduced for the ME process by techniques that provide a scalable matching criterion on the vector-selection level and/or the accuracy level [8], [9]. In [10], a structural-level ME technique is presented, which processes MVFs twice for quality improvements. In this paper, we present two new techniques to insert complexity scalability into the ME process.…”