PRSD Studio

Features

Fast execution of pattern recognition algorithms using an optimized C library (in Matlab as a MEX and out of Matlab using a DLL). learn more...
Easy-to-create custom algorithms.learn more...
Two-class and multi-class ROC analysis. learn more...
Decisions at any user-defined operating point.
Seamless integration with the PRTools toolbox. learn more...
Easy integration into any custom application which can call a DLL (no dependency on Matlab).
No re-compilation or re-linking needed to update a production application with a re-trained or entirely newly designed recognition pipeline.
Transparent multi-core support.
Cross-platform (Windows, Linux, Mac OS X).

Linear algorithms: PCA, nearest mean classifier, Fisher linear discriminant, Logistic linear classifier, Perceptron.
Normal-based algorithms: Linear, quadratic, mixture of Gaussians.
Non-parametric: Parzen classifier, k-NN (nearest neighbor).
Support Vector Machines: Polynomial and RBF kernels.
Classifiers trained in a dissimilarity space (i.e. Fisher on distances to prototypes).
Decision trees
Output normalization using the sigmoid mapping (Useful for classifier fusion and for building multi-stage non-linear systems using linear base classifiers).
Decisions using the target-thresholding and output weighting (operating points may be adjusted from the custom application.).
Trained combiners (fusion of classifier outputs by a trained rule).
Any combination of the algorithms above using:
- Sequences (i.e. PCA followed by a mixture of Gaussians).
- Stacking (i.e. combination of multiple classifiers in the same feature space).
- Parallel combination (different classifiers in different feature spaces).
- Cascading (i.e. discriminant executed only on samples labeled as target by the detector).