PDF version of this document

Shapes Revisited

Upon returning to the subject of shapes and automatic landmark selection, goals could be laid out perfectly well. There were the issues of speed and instability of the objective function. Also, it was vital to make the code work in the absence of its original developers and authors. These problems were all shortly embarked upon and some successful solution were found. Figure shows the data which was typically handled^7.20, namely the brick-and-bump data. The 3 principal modes of variation are shown for the raw data at the start when points are spread at equally-spaced locations along the curve.

Figure: The unregistered bump data and its three principal modes of variation (2 standard deviations).

To ease the operation of the code, an interface (Figure ) was built and used to bring the code back into workable order. The process otherwise involved manipulation of the code which requires the user to know the algorithm. This situation proved to be difficult and tolerable at best.

Figure: graphical user interface for semi- automatic landmark selection as of June 2004.