Pattern recognition algorithms in SPRLIB

Here follows a brief list of the pattern recognition methods implemented in the pattern recognition part of the library. The reader is reffered to the existing literature on pattern recognition for the details of the different functions, or to the reference part for a global explanation.

• Mean and covariance estimation (mean_var_dataset)
• Within and between scatter estimation (all_scatter_dataset)
• Principal component analysis (kl_coordinates)
• Random number generators (e.g. random_multi_init)
• Fisher linear discriminant function (e.g. fisher)
• K-nearest neighbour classifier (e.g. knn_classify)
• Nearest neighbour classifier (e.g. nn_nearest_neighbour)
• Clustering methods (e.g. isodata_basic)
• Parzen classification (e.g. parzen_classify)
• Quadratic discriminant function (e.g. fisherq)
• Multiclass Mahalanobis classifier (e.g. mahal_multiclass_dataset)
• Error estimation routines (e.g. calc_classerror)

Note that most pattern recognition methods require more than one function call. This means that in most cases first an initialization function has to be called before the actual method (read function) is applied. For instance when using the k-nearest neighbour method first the function knn_init must be called, after which the different samples can be classified using knn_classify. Finally when having done all classifications the function knn_free) has to be executed to free all the memory structures which were needed for the k-nearest neighbour classifier.