Video compression has become an important aspect in the current drive for digital video technology.
Abstract
Video compression has become an important aspect in the current drive for digital video technology. The success of many emerging applications such video on demand, teleconferencing, digital libraries etc. rely critically on improvements in video compressions.
However, most compression techniques such as motion detection require a great deal of computation resources and specialized hardware which are costly.
This project demonstrates a high performance compression system based on 3D cosine transform coding, that its computation needs are not so demanding and therefore can be implemented in software.
The problem
Although the added value of relatively low computational demands of the 3D-DCT, its compression ratio lacks performance. Ways are needed to improve on the bandwidth requirements without hindering the visual quality of the decoded movie.
The solution
This project will attempt to improve and the bandwidth requirements by adopting a special quantization method developed by Raymond, Chan and Adjeroh, and applying several methods that have been proved worthwhile in image processing.
Given a package of 8 frames, it is divided into 8*8*8 3D blocks. Each block is transformed and quantized.
Then quantization values where derived using several techniques:
Applying a band pass filter that matches the human visual system.
Applying a rough quantization to texture areas in the movie that are hardly noticeable to the human eye but are costly in compression.
Applying a rough quantization to the DC coefficient.
In addition blocks that are almost identical to their previous counterparts are not transmitted again.
Block diagram of the system/algorithm
Tools
We developed 2 applications:
A simulator module to generate the output movie.
A viewer to inspect the visual quality of the resulting movies.
Conclusions
The compression ratio was improved considerably without affecting the visual quality (See attached histogram).
Input/output of the system
The system receives a movie in CIF format and produces a reconstructed movie and the compression statistics.
Project Results
The histogram below demonstrates the improvement in the compression ratio of each technique relative to the basic 3D-DCT:
It can be seen that the most significant improvement was due to applying the HVS filter and block matching.
The final compression ration reached 160, while basic 3D-DCT achieved only 91.
The movie’s visual quality remained unchanged.
Acknowledgments
We would like to thank our supervisor Dmitry Furman for his support and guidance throughout this project.
Also we would like to thank the Ollendorff Minerav Center Fund which supported this project.