Andrew Lewis
Part II projects
Student-suggested projects
I'm interested in supervising projects related to
- Image and video compression (JPEG, JPEG-2000, MPEG-1/2/4, H.264/AVC, etc.)
- Watermarking and steganography
- Coding and information theory
Lossless editing for H.264 video streams
Consumer video editing software operates on the decompressed representation of a compressed video bit-stream produced by a digital camera, and recompresses the resulting edited video when the user is finished. For lossy compression schemes, the final recompression causes unnecessary information loss in some cases.
This is an open-ended project which will involve investigating the feasibility of applying common video editing functions directly to a compressed H.264 bit-stream, to avoid information loss and speed up the final export operation. In particular, it may be possible to perform the following editing tasks directly on the compressed bit-stream in certain circumstances:
- cutting operations which remove or re-order scenes in the movie;
- stabilization of video content;
- conversion ('transcoding') between different profiles/levels, to make video suitable for different devices or services; and
- efficient production of video thumbnails.
The project might involve writing a rudimentary graphical user interface or automatic command-line tool so that users can easily apply these operations, or see under what circumstances they can be used.
Container-independent parsing of H.264 video
Although the binary format for H.264 format is precisely specified, engineers may embed the components of the bit-stream in one of many different container/transport formats, or use their own proprietary container. This presents a problem for organizations which receive video from a variety of sources.
This project involves writing a tool which inspects a binary file provided as input and extracts as much information as possible about the video, regardless of the details of the container format.
Image/video forensics
Since 2004, many academic papers have been published which propose algorithms that use signal processing techniques to detect and characterise tampering in digital media, and gather information about the capturing hardware and processing history of images, video and audio.
- source camera identification uses intrinsic non-uniformities in the CCD chips used in digital cameras to determine which particular camera was used to capture an image;
- copy/move detection techniques indicate regions of an image which are very similar or identical to other regions, hopefully detecting careless use of a photograph editor's clone tool;
- double compression detection algorithms look for indications of recompression in the distribution of discrete cosine transform (DCT) coefficients in JPEG images, which indicate that an original compressed image may have been tampered with before recompression;
- resampling detection finds regions where the inter-pixel correlations are consistent with interpolation, indicating that the areas contain content which was probably resized or rotated.
My multimedia forensics bibliography includes links to papers tackling a variety of problems in multimedia forensics.
This project was taken by Greg Borek in 2008/09
Watermarking Dirac video streams
This project was suggested and taken by Andrew Jones in 2008/09