PDF version of this document

next up previous contents index
Next: Description of Task Up: INTRODUCTION Previous: INTRODUCTION   Contents   Index


Project Background

\includegraphics[%%
scale=0.3]{./Graphics/a.eps} suitable way of describing this undertaken research is by briefly describing its aims in a simplistic form that requires limited understanding of its background. Context is key to the understanding of how current knowledge pertains to and contributes to the main hypothesis.

Given a collection of data objects (quite commonly in the form of two dimensional images) which are clearly different although they describe the same object, one wishes to transform them in some way so that they appear as identical as possible to one another. The solution to this task cannot be unique, meaning that there will be infinitely many solutions, i.e. transformations, that get similar results. For example, common sense may suggest that two such data objects should be selected each time and, subsequently, one of these objects should be transformed to fit the other. This raises the questions: Which objects should be selected? How should they be transformed? What conditions define a good fit?

One can use an existing technology to model all of these images and use this modelling process to minimise a term of complexity. The basic contention is that when this term is minimised, better identity across the set of data is granted and a single unique solution is always reached.

This process above is very beneficiary because one if its byproducts is a description of a group of transformations - the transformations that were used to manipulate data to attain identity. Such descriptions can be used, in a process of learning, to form knowledge about the observed differences in the data set. They can describe how to transform a single data object in a way which preserves prevalent, well-ground variations. They can be used to construct models which are capable of regenerating existing and yet unseen data that exhibits similar properties.

For raw data (as described above) to be modelled properly, knowledge about corresponding patterns and points in the data, must be gained. Thus far, human understanding of the data aided a process of annotation. That process involved mark-up of data regions or points that are homologous. However, once data is made merely identical, mark-up becomes trivial. This is because points tend to lie at identical position. Significantly enough, no manual mark-up of the data is necessary once the approach outlined above successfully works. The questions that remain are: Can data sets be transformed to reach a state of identity? Will the framework of models transcend the peril of data being changed?


next up previous contents index
Next: Description of Task Up: INTRODUCTION Previous: INTRODUCTION   Contents   Index
2004-08-02