Richard Wilson's project proposals (2008)

RCW/1 
Project 1: Handwritten Postcode Recognition [CS,CS/M]

Aims: The aims of this project are

Background: Many millions of letters are sent through the post each day. Business letters almost always have printed addresses and are relatively easy to read automatically using OCR (Optical Character Recognition). Handwritten letters however present a more challenging problem because of the huge variation in handwritten. This project will focus on a subproblem; reading handwritten postcodes. 

Handwriting recognition is generally split into two stages; segmentation, where the writing is split into individual character images, and recognition, where the character images are converted into characters. In this project you will be expected to implement segmentation and recognition on postcodes.

Reading

Some concepts of character recognition, A. W. Holt, 1976

Matteson, R (1995). Introduction to Document Image Processing Techniques. Artech House.

Trier, Ø. Jain, A. Taxt, T. (1995). Feature Extraction Methods for Character Recognition – A Survey. Pattern Recognition. Vol 29. No. 4. pp 641-662

RCW/2 
Project 2: Structure from motion [CS,CS/M]

Aims: The aim of this project is to reconstruct the shape of an object from an sequence of images where the object (or camera) is moving. Background and method: It is well known that the shape of an object can be found using a pair of stereo images of the same object, but from different locations. It is also possible to find the shape of an object from images taken at different times, provided that either the object or the camera is moving. Structure from motion solves this problem. The solution is divided into three parts; 1) finding the optical flow between images, 2) calibrating the camera and motion, and 3) finding the shape. In this project, you will implement methods to tackle these problems and extract the shape of an object.

Reading

Richard Hartley and Andrew Zisserman (2003). Multiple View Geometry in Computer Vision. Cambridge University Press.

T. Huang and A. Netravali. Motion and structure from feature correspondences: A review. Proceedings of the IEEE, 82(2), 1994.

RCW/3 
Project 3: Graph descriptors [CS,MEng,CS/M,MMath]

Aims: The aim of this project are to design and test the effectiveness of graph descriptors for uniquely distinguishing between small graphs. You will need to do the following Background and method: Graphs are difficult to compare to each other, since there is no known polynomial-time algorithm for deciding if two graphs are the same (isomorphic). A graph descriptor is a number (or set of numbers) which can be used to represent a graph and can be computed relatively easily. An example is the number of edges in the graph. The idea is that, if two graphs are the same, then the descriptor should be the same, and if two graphs are different, then the descriptor should be different. There is no known descriptor which satisfies this property, as it would solve the isomorphism problem above, but some are better than others in the sense that they often uniquely describe the graph, but not always. In this project you will design and test descriptors to see how good they are on small graphs. M-level students will be expected to design novel descriptors and test their effectiveness.

Reading

P. Zhu and R. C. Wilson. Stability of the eigenvalues of graphs. Computer Analysis Of Images And Patterns, Proceedings, 3691:371-378, 2005.

R. C. Wilson, E. R. Hancock, and B. Luo. Pattern vectors from algebraic graph theory. IEEE Transactions On Pattern Analysis And Machine Intelligence, 27(7):1112-1124, July 2005

Willem H. Haemers, Edward Spence: Enumeration of cospectral graphs. Eur. J. Comb. 25(2): 199-211 (2004)


RCW/4
Project 4:  Identifying Galaxies [CS,MEng,CS/M,MMath]

Background and Method: Modern telescopes return an enormous about of image data. Some of these are 'deep fields' which contain images of many thousands of galaxies. One of the things that astronomers would like to do is classify and count these galaxies into different types, as this gives some important insights into the history of the Universe. In fact there is an online project in which people can participate by identifying galaxies by eye (www.galaxyzoo.org). The aim of this project is to design and implement a method of automatically classifying them. You will need to investigate and design image features which will distinguish the different types and build a classifier to sort them into the different types. At M-level you will need to create new descriptors and compare their performance against some of the alternatives in the literature.

Reading

www.galaxyzoo.org

http://www.sdss.org/

Identification and classification of galaxies using a biologically-inspired neutral network, Radhakhrishna Somanah, Soonil D.D.V. Rughooputh and Harry C.S. Rughooputh, Astrophysics and Space Science, 282(1) pp161-169, 2002

RCW/5
Project 5:  Musical Genre Classification [CS,CS/M,MEng,MMath,MScIT]

Aims: The aim of this project is to design features and a classifier which will sort music into a number of different genres. The main tasks are Background and Method: Sorting music into different genres is an extremely useful task when organising a large database of songs. While the classification of music is somewhat subjective, there are features which can distinguish different styles. For example, instrumental music can be distinguished from songs by looking for stable frequencies in the spectrogram. In this project you will look for similar features in the time and frequency domain which can distinguish more different musical genres or, for an M-level student, you will need to devise a new method of analysis which can separate the different genres.

http://www.hpl.hp.com/techreports/2003/HPL-2003-183.pdf

Automatic Music Classification and Summarization, C Xu, N C Maddage and X. Shao, IEEE Transactions on Speech and Audio Processing, 13(3) pp441-450 2005

RCW/6
Project 6:

Shot Segmentation in Videos [CS,MEng,CS/M,MMath]

In order to produce useful video libraries, it is often important to split a video sequence up into it's shots or scenes. Each shot is taken from one camera, and each scene is set in one specific location. The aim of this project is to devise a method for segmenting a video clip using either the geometry of the image for shots, or features of the location for scenes. There are a number of features of video frames which can be used to distinguish shots, ranging from colour histograms and texture analysis to speed of motion. In this project, the student will either implement such a method or, for an M-level project, devise a new method and apply it to segmenting shots in short video clips.

Reading

 http://www.nada.kth.se/cvap/VIBES/

Handbook of image and video processing,  Al Bovik(ed), San Diego : Academic Press, 2000

Computer Vision, Ballard and Brown, Prentice Hall

Computer and Robot Vision, Haralick and Shapiro, Addison Wesley