Potential Projects

Below I provide a list of two indicative MSc projects of interest to me. Please email me if you would like to discuss these further. Simplified versions of these projects can be undertaken for a 3rd year or 4th year student project.

MSc Project 1:

"Comparison of Incremental Salient Point Detectors: Theory and Application"

Description of the Work:

Corner and edge detection or the more general terminology “interest point” or “salient point” detection is an approach used within computer vision systems to extract certain kinds of features and infer the contents of an image. Various applications can use such systems for image analysis, image and video coding, pattern recognition, etc. To cater for image regions containing texture and isolated features, a combined corner and edge detector based on the local autocorrelation function is commonly utilized.
In this work we propose to use the recently-proposed modified version of two of the most popular salient-point detectors in the literature that operate in an incremental manner. Specifically, we assume a coupling of the image sensor with the image processing system that provides individual bitplanes of the image from the sensor to the image processor. The modified version of the algorithm can use these “increments” of the image information to successively refine the computation of the detector results. This means that the results can be refined with additional computational effort instead of re-computing everything from scratch.

This work will provide an extensive evaluation of the results of these incremental salient point detector algorithms versus the original algorithms that are not refinable and compute the detector using all the available information at each instance. Theoretical and practical extensions of the algorithms will be attempted and evaluated. Applications in visual sensor networks and analysis of visual information will be analyzed.

Prerequisites: Good knowledge of C programming, at least one course in Signal Processing and/or Image Processing, familiarity with the Matlab programming environment.

MSc Project 2:

"Analysis of Adaptive Transform Decomposition Systems with Noise: Applications in Video Communications via Unreliable Networks"

Description of the Work:

All the popular image and video coding standards such as MPEG-2, MPEG-4, AVC, JPEG and JPEG-2000, use some form of transform decomposition (e.g. the Discrete Cosine Transform) and motion estimation in order to efficiently decompose the input image or video to certain spatio-temporal frequency bands that are easily compressible.

The new trends for future and emerging standards for multimedia coding and communications use adaptive transform decompositions, where the transform itself is modified according to signal properties, such as sudden changes in motion, or sudden illumination changes in images. In typical application scenarios, the produced information is compressed and transmitted via unreliable networks (e.g. Wireless IEEE802.11a networks). There, noise is potentially inserted in the transform coefficients that the decoder will use in order to reconstruct the multimedia information at the receiver side.

In this work we propose to study this process using classical results from perturbation theory of linear systems. The impact of the noise can be expressed analytically for the reconstruction process and bounds can be derived for the expected reconstruction error. These bounds translate to estimates for the visual distortion at the video receiver. Popular systems for video communications will be studied from the literature and operational algorithms will be developed for control of the adaptive decomposition based on the derived reconstruction-error bounds. Applications will be studied in the areas of video streaming quality optimization, and rate vs. coding-plus-channel-distortion optimization for video coding.

Prerequisites: Good knowledge of C programming, at least one course in Signal Processing and/or Image Processing, familiarity with the Matlab programming environment.

This page last modified 30 January, 2008 by [John Mitchell]