PDF version of this document
Past work by Kotcheff and Taylor [] attempted to address the problem which was highlighted earlier on. This problem is that of selecting good corresponding points on a curve of shape. It did so by evaluating the choice of landmarks5.1 via the model which resulted from that choice. The determinant of the covariance matrix of the model was said to be a good approximation for model quality. Poor models, it is yet again worth emphasising, imply that when PCA is applied to the data, the correctness of corresponding points and their distribution, will be invalid. Some of the technicalities will be described later on, along with mathematical notation.