Scientific documents are particularly difficult to recognise with OCR software. They contain diagrams, graphs, mathematical formulae and complex tables. Because of the difficulty of analysing them, students with visual impairments are much more severely disadvantaged in studying the sciences than in studying many other subjects. This set of projects (there are many different ones in this area) is to use a particular scanned book (http://www.cs.bham.ac.uk/~aps/research/projects/as/) in developing tools, algorithms and solutions to assist in this research.

Possible projects include: developing a character recogniser for the individual characters in the image. Analysing the layout of the page, to identify blocks of the pages as containing columns, headings, diagrams, formulae etc. Recognising tables or formulae. Turning bitmap diagrams into vector diagrams etc.

Online methods of optical character, formula and diagram recognition, online recognition are based on drawing on a pad device, a graphics tablet, a smartphone touch screen or with a mouse. Each stroke that that the user enters is analysed, recognised, corrected and displayed to the user, who can either accept what has been recognised or undo the change. Demonstrations of a commercial system that implements a number of tools of this type are available at: (http://webdemo.visionobjects.com/)

This project is to focus on one of these application domains (text, formulae or one diagram type, e.g.UML Class diagrams, Chemical diagrams, etc.) and implement a working system.

The system can be a Web application, using HTML5's Canvas element, CSS and Javascript for the GUI front end, it could be a smartphone application (Android preferred), or it could be a desktop application using Java or OCaml. Also possible is the use of a note taking device such as the Digimemo: http://www.digimemo.net/

Recognising printed characters correctly from scanned images is difficult because of the range of ways in which the scanned image may deviate from the true image. This project is to choose one of a number of different technologies and develop an OCR engine to recognise characters. Suitable technologies from image analysis and pattern classification and clustering can be used.

Historical manuscripts contain handwritten text that, in general, is beyond the current ability of handwriting analysis systems to recognise. Hence for many such manuscripts, humans read the manuscripts and key in the corresponding text so that various search tools can index and find the corresponding pages. This is a laborious and errorprone task, not least because checking for errors in the keyed in text requires matching the written text with the keyed in text. This project is to assist in this task by trying to carry out that matching automatically: this is a much easier task than handwriting analysis but would be extremely useful to a large number of researchers in the Humanities. It would be most useful to have this tool as a rich GUI web application, although a desktop application would also be acceptable.

Archaeologists and Historians, in their studies of ancient languages, manuscripts and writing, often need to do very difficult and painstaking character recognition of symbols and text. Recently some illuminated manuscripts from the 11th and 12th century have been scanned at very high resolution (c.f. http://www.isos.dcu.ie/) and are available for inspection and processing. While the languages of these manuscripts may not be known to the students (indeed, in some cases the interest is to decipher lost languages), computer techniques to match images, extract common symbols and cluster patterns of images, together with various image enhancement techniques, could prove very useful.

I have obtained a large collect of images of manuscripts of the the Pilgrims Progress. There is significant scientific interest in determining information about the scribes who wrote them. I believe that image analysis techniques should be able to find objective evidence supporting or contradicting commonly held beliefs about the origin and development of these documents

This project is to develop a collection of tools in a software toolbox useful for such purposes. It will involve development of specialist noise reduction, deskewing, binarisation, glyph extraction, shape clustering and identification algorithms.

Drawing diagrams for inclusion in papers and reports can be very tedious. Either you use a markup language for drawing figures which gives a great deal of flexibility in defining your diagrams but needs programming style skills to do so, or you use an interactive tool and, typically, sacrifice some expressivity or quality of the final diagram. Either way, it can take a long time and a great deal of detailed work to get it right.

An alternative is to sketch the diagram on paper, scan it in, analyse the results for lines, curves, rectangles, circles, text etc. use some intelligence to clean it up, e.g. by guessing that, a particular slightly wavy line was intended to be straight, and generate the diagram in an appropriate form for further editing in a diagram tool.

Even the best commercial OCR programs can rarely do more than recognise diagrams in documents as images. However, it would be nice to be able to analyse the image and turn it into an editable vectorised image that could then be further modified with a suitable diagram editing tool. Furthermore, the result should require far less disk space than the original image and could be printed at much higher resolution than the original (without showing the grainy quality that printing bitmap images at too high resolution incurs).

This requires scanning the diagram, analysing the results for lines, curves, rectangles, circles, text etc. use some intelligence to clean it up, appropriate form for further editing in a diagram tool.

This project is to optically scan a sheet of music and analyse it. Minimally, a Music XML file should be generated that captures as much as possible of the music information on the music sheet. Generating MIDI output from the Music XML file would be a useful bonus.

The student should understand at least something about reading music and music theory although he/she does not have to be an expert.

Using a tool to extract characters from PDF documents that will be provided to the project, a mathematical expression can be scanned resulting in a list of which characters are in the image, where they are in relation to each other (in 2 dimensions) and what size they are. The task is to construct the mathematical expression including subscripts, superscripts, fractions, summations, matrix expressions and as many other mathematical expression components as possible.

One of the most difficult problems in OCR is dealing with different glyphs that touch each other or damaged glyphs. While we can get very high recognition for individual glyphs, when two glyphs actually touch, or due to printing problems, a single glyph is broken into two separate components, character recognition rates drop significantly. This project is to explore this problem, implement and evaluate some standard approaches described in the literature, and design and develop new algorithms based on what you have learned.

One the the most promising techniques for higher level analysis of documents (i.e. after all the low level image analysis and character recognition has been completed) is Graph Rewriting. The idea is that the objects discovered in an image are placed in a graph data structure with links between objects based on their spatial relationship to each other (e.g. to-the-right-near or above-left-far etc). Identifying the higher level structure in the document can then be done by rewriting the graph: e.g. if one object is an italic "x" at ten point size and above-right-near there is a "2" at 8 point size we might reasonably rewrite the two objects and the link into a single "x squared" object, while keeping the remainder of the graph the same. A whole sequence of rewrites could recognise a whole mathematical expression. The same techniques can be applied to diagram recognition, form recognition or general document analysis.

This project is to develop a working graph rewriter and associated algorithms to experiment with these ideas.

Paper forms such as the school's mid and end semester questionnaires require a great deal of work to prepare and expensive software to analyse them. This project is to produce a free, public domain version that could be the basis of a cheap but effective solution for anyone who needs to work with forms. There are many different components of a scanner based forms processing system. The forms have to be created. This can be in a special purpose editor which allows creating the various components: choice fields, image fields, text boxes etc. Alternatively it can be produced outside the system and then scanned in, allowing a user to mark up and configure the various components. Various image processing facilities are necessary (noise reduction, skew and rotation correction etc). Automated form detection is an advantage, locating components correctly on the form essential, segmentation issues must be looked at etc. Good optical character recognition on top of the rest would be too ambitious for this project but the system should provide an API by which an OCR engine could be plugged in to the system and a free open source OCR engine such as Tesseract or OCROpus could be used.

Ocsigen is an OCaml based web application framework with some interesting and very attractive features for easy development of sophisticated web applications. This project is to extend those features to the new HTML5 canvas elements to provide a framework suitable for rich graphical web applications. The project will involve building a demonstrator application that shows the new features in operation.

Graphs (the node/edge variety) are used in many areas of computer science both as internal data structures and as interface elements. As a result there is often a need for displaying a graph in an aesthetic manner. Unfortunately it is very difficult to come up with a good algorithm for doing so although there is now a rich literature on the problem.

This project is to review the literature and implement a number of the most interesting graph layout algorithms in a useful environment: namely, the graph layout algorithms should be implemented as a module which can handle annotating a graph with appropriate spatial coordinates. A separate widget module should then see to display of the resulting graph with support for selecting nodes and navigating the graph structure. The entire package then becomes a useful interface tool for inclusion with any program that has a use for graph or tree representations in its interface.

Graph algorithms (e.g. shortest paths, finding minimal cycles etc) are very important for many applications, especially for those with more sophisticated requirements, and yet are poorly understood by many programmers. OCaml is an excellent language for working with graphs. This project is to develop implementations of standard graph algorithms and demonstrate how they work with graphical animations.

The game of Go has stumped computer scientists for many years. Computer programs are now the best players of Chess, Othello/Reversi and (bar one human) draughts/checkers. However, Computer Go players have still not advanced beyond the level of amateurs. It would not be feasible to try to improve on this in a final year project, but some parts of the overall problem could be tackled. For example, "Life and Death" problems in Go are published in news papers and are equivalent to the Chess problems of "White to mate in 4" and the like. This project is to develop a program to visualise and solve such problems. The student should already be familiar with the game.

A common tool for organising ones thoughts, and a favourite tool of students planning essays, is "Mind Mapping", typically involving taking a large sheet of paper and drawing lots of cloud diagrams with lines between them. There are a number of desktop applications to support such a process (e.g. Freemind). This project is to develop a web based version that allows groups of people to develop mind maps together online.

For someone with a large number of books, keeping track of them can be a problem but necessary for reasons of insurance, security in lending them, and just assistance in finding them. Entering the data alone is such a nuisance that many people simply don't bother. Now, one can obtain the full bibliographic data from the ISBN information (which can be entered by hand or which can be scanned with the aid of a cheap bar code reader). There are desktop applications for this purpose. However, it would still be nice to have full access to the book collection manager online, so that one can enter new books at home, at work or when away on holidays or at conferences. Similarly one would like to be able to carry out all possible operations on the collection online rather than be fixed to one location for the purpose. This project is to do precisely that. Other functionality, such as generating bibliography information for inclusion in academic papers and reports, or supporting other forms of publications, such as papers, music or videos is also possible.

Family trees are rarely trees. With possibilities of deaths, divorce, remarriage, adoption etc., a family tree can be quite a complex graph. Furthermore, it is often very hard to collect together the family history from an extended family, each member of which may know some small bits that no-one else knows, or has some pictures or anecdotes that are not available to the others.

This project is to develop a web application that securely allows a family to collaborate in collecting together and sharing the family history and relationships. Family members should be able to log in, see various visualisations of the family structure, add pictures, video or sound recording, or text annotations.

The problem here is to take a MIDI file and analyse it to extract the music information. The music information should be output as sheet music.

The student should understand at least something about reading music and music theory although he/she does not have to be an expert.

An excellent site on music analysis can be found at OMRAS (Online Music Recognition and Searching)

I have supervised a number of music related projects in the past, and one significant problem with all of them is the lack of a freely available high quality user interface for dynamically displaying and interacting with a graphical representation of sheet music. Think of it as the musical equivalent of a Java Swing JEditorPane: One should be able to display music scores in it, dynamically modify the contents of the pane, highlight parts of the score, etc. Such a Swing control could be used as the front end to any number of music related applications: from capturing midi on a keyboard and display the corresponding music in real time, to playing music while highlighting on the score exactly what is being played, to providing the front end to a music score editor.

The student should understand at least something about reading music and music theory although he/she does not have to be an expert.

UML contains 14 different types of diagrams and I would like projects to develop editors for some or all of these diagrams. Support for intelligent editing will be required (e.g. the inclusion of top-down and bottom up editing operations on diagrams), as well as output to SVG, XML and bitmap image formats.

There is scope to implement this either as a desktop application or as a web application. As a web application, there is the possibility of making it a communal application where multiple people can work on the same diagram simultaneously.

Entity-Relationship-Attribute modelling is the most common form of conceptual modelling used in database design today. Tools to support ERA modelling are available but are, for the most part, either very expensive or of poor quality. An ideal ERA diagram drawing tool is more than just a figure drawing tool with some special predefined figures. It should support common refinement approaches (Top Down, Bottom up, Inside Out and Mixed). It should support different diagramming styles such as Chen (rectangle/diamond/oval) and Crow's Feet. It should be able to save to and restore from the Conceptual Schema Definition Language (CSDL). It should support view integration, constraints and extended ERA features such as Generalisation, Aggregation and Classification. It should be able to map a conceptual model into a logical model (e.g. a relational schema). It should be able to import a relational schema and make reasonable assumptions to extract an approximate conceptual model from it.

This project is to design and develop an ERA tool that goes as far as possible in producing such a system. The internal design and user interface design are critical parts of the project as the intention is to carry the project further in later years.