An enhanced estimation: motion and rotation estimation SHAN-HSIANG SHEN Motivation Video encoding methods such as H.264 and MPEG use motion estimation to increase compression rate. Paste each blocks from some reference frame to create current frame. The more pixels can be matched, the more frame quality it can reach. Not only motion, but rotation. Motion estimation Motion: Rotation: Rotation angles How many angle we will estimate depends on the block size. Evaluation Evaluation program with Java Media Framework (JMF) Motion and Video Rotation Video reader estimation Writer Motion vector and angle (SAD for each frame will be available here for evaluation) Evaluation Test video : 4.24 second-short video with 108 frames including some rotation object. Total SAD for each frame shows difference. 18000000 16000000 14000000 12000000 Without rotation With rotation 10000000 Total SAD 8000000 6000000 Average SAD without rotation: 4000000 3812728 2000000 Average SAD with rotation: 0 3681778 0 20 40 60 80 100 120 -2000000 sequence number of frames Conclusion Motion and rotation estimation increase the number of pixel can be matched between current frame and reference frame. SAD is improved but not so much. The number of angle we estimation will be a trade- off between encoding time and performance.
Pages to are hidden for
"An enhanced estimation: motion and rotation estimation"Please download to view full document