Design And Implementation Of Fast Motion Estimation In Modern Video Compression On GPU

<p>Motion estimation is the most compute expensive part of high definition video compression. It accounts for more than 50\% of overall execution. Therefore, improving the performance of motion estimation can make significant impact on the overall performance of video compression. The performance of motion estimation can be improved in two aspects: algorithm and implementation. This thesis touches both aspects. We first propose an innovative motion estimation algorithm by replacing the traditional block matching method which comparing blocks pixel by pixel with a brand new method which based on lbp (local binary pattern) code. Our new method first encodes the original video frames into lbp code and then compares the blocks only using the lbp code. Our algorithm reduces the amount of computation significantly by avoiding many pixel by pixel comparisons present in traditional block matching approaches. Using public benchmarks our experiments show our proposed motion estimation algorithm runs 5 times faster than a traditional algorithm. Furthermore, we accelerate our proposed algorithm on gpus. Motion estimation processes of all blocks are offloaded to gpu and accelerated in parallel. Our gpu implementation runs 9 times faster than cpu implementation.</p>

Name thesis:degree_name

M.S. in Engineering Science

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer and Information Science

Year dc:date.available

2015

Author dc:creator

• Yi, Zhaohua

Contributors dc:contributor

• Byunghyun Jang
• Feng Wang
• Yixin Chen

• Gpu Computing
• Motion Estimation
• Computer Sciences

Repository record dc:identifier

https://egrove.olemiss.edu/etd/949

OAI identifier oai:identifier

oai:egrove.olemiss.edu:etd-1948