Recasting a Scene Adaptive Video Coder for Real Time Implementation

The Scene Adaptive Video Coder is a new algorithm used to code video sequences at very low bitrates (8 kb/s to 40 kb/s). It is a constant bitrate, constant frame rate encoder. Low rates are obtained through the use of sub-pel motion estimation, aggressive adaptive vector quantization, and dynamic bit allocation. Unfortunately, the algorithm is too complex to be implemented in real time, which is important for practical applications of low bitrate video. This thesis develops a recast of the algorithm to reduce its computational complexity while maintaining its important features, and minimizing the performance degradation as compared to the original algorithm. The computational trouble spots of the algorithm are determined, models are developed to assess the computational complexity and performance trade-offs, and the newly designed algorithm is implemented. Thesis Supervisor: Dr. John Hartung Title: Member of Technical Staff, AT&T Bell Laboratories Thesis Supervisor: Dr. Jae Lim Title: Professor, MIT Acknowledgments I would like to thank my thesis advisors, Dr. John Hartung and Prof. Jae Lim, for their guidance and support throughout this project. I would also like to thank Dr. Nikil Jayant, and the rest of the staff in the Signal Processing Research Department, for their assistance. A special thanks go to Dr. Rich Pauls, for many technical and philosophical discussions. Lastly, I would like to thank my fiance, Michelle Greene, for her continual support of my work and her patience with me. My parents also deserve a special thanks for their constant support throughout my education.