Video Coding: Comparison of Block Matching Techniques In Motion Estimation

This project is to investigate the advantages of using the various types of block matching
methods as a type of motion estimation techniques compared to encoding each frame as a
separate static image. As video is continuous media, it is important to maintain its quality
and efficiency while compressing it. As the similarity of a frame and the next frame is
great, it can be used as the advantage in video coding. This is because, the background of
the image will usually stay the same and the only thing that will change is the moving
object in that video. In order to achieve this, the author has to develop a program that can
accommodate motion compensation and estimation by utilizing the computational for the
motion estimation techniques, if applicable. The author will then have to run the program
with some test video sequences to compare the performances of different block matching
techniques for different types of video sequences. The major types of block matching
techniques are Full (Exhaustive) Search, and a Fast Search (Three Step Search) has been
chosen for this project. The author has also chosen to work on Quarter Common
Intermediate Format (QCIF) video sequences. As the purpose of this project is to
investigate the motion estimation technique, only the video frames will be considered and
the sound of the actual video will be left out.
Experimental results show that MMSE has a better PSNR value than MAD but consume
more time and has higher complexity of operation. Block sizes and window sizes also
have a significant effect on the predicted image. The Three Step Search has experiments
has shown that it has a higher speed ratio as compared to Full Search, but with reduced
quality.