PDF version of this entire document

Image Data

It seemed like a reasonable choice to concentrate on images which serve as simple test data, so an algorithm was constructed which synthesises elliptic `bumps'. Initial experiments on MDL image registration were performed using one dimensional synthetic images. Examples of the synthetic bump images are shown in Figure where the sort of variation permitted in the generated data is demonstrated.

The data is generated by moving along $\mathbf{x}$ (from the left side of the image to the right) and considering 3 random variables, $\{p,w,h\}$, which lie within a sensible range that restricts variability. The variables correspond to:

1. The point where the bump starts;
2. The width of the bump (which defines where the curve flattens and the bump ends) and
3. The height of the bump.

These variables are referred to as $p$, $w$, and $h$, respectively.

Figure: Examples of image data which illustrate of the three variation modes, $\{p,w,h\}$. These variations characterise the data to be analysed. It is not the same image being resampled; rather, these are separate images.

We refer to the height of the curves as image intensity or texture, whilst referring to the position along the x axis of particular features (e.g. the start and end of the bump) as shape.