In order to generate a new template model, construct a simple asci file (``filename''.blt) containing the file names of the required profiles. Specify the name of this file in the filename parameter. The model is built using the PCA routine and is automatically output to the filename model files on completion.
The eigen values for the shape model will be written to the main tinatool dialog window and can be used as a guide for subsequent selection of the number of principle modes. As scale does not appear as a mode of variation zero modes of variation corresponds to scale change only. At the same time the mean shape and profile model is written to the selected single model file. The model builder will use the options Normlise and Global if selected.
The Normalise option is intended for image data which has variable greylevel scale.
Use of a Global correlation model between both shape and grey-level parameters. This model choice has considerable computational overhead and should only be selected for problems where there is a strong correlation between grey level profile and shape and the location problem proves to be particularly challenging. Accurate determination of the extra covarinace parameters also requires significant increases in the size of the training data set.
The file (``filename''.pca) which stores the parameters of a specific trained model are specified by the filename parameter string. The mean model (``polyname''.mml) can be loaded from polyname using input. This flexibility is provided so that the allowed variation specified by the PCA parameters for one dataset can be quickly applied to another.