Login Register

Auto-login on future visits

Forgot your password?

Newsletter July-August 2010

How to protect trained classifiers from outliers?

Tutorial Outliers protection

Classifiers we train are often executed in environments where new types of measurements appear that were not considered during classifier design. For example, in a fruit sorting problem, our classifier distinguishing several types of fruit may also encounter stones, leaves or dirt on the conveyor belt. Accepting stones or dirt as one of the fruit classes results in high sorting error.

In this tutorial, we discuss how to protect the trained multi-class discriminants from accepting such outliers.

See the video in our blog.

Visit us in Istanbul at ICPR 2010!


Our latest research on ROC analysis for hierarchical classifier will be presented at ICPR (International Conference on Patter Recognition) in Istanbul, Turkey (23-26 August 2010). Our poster session is on Wednesday afternoon [13:30-16:30]. We will be happy to meet you! See the full ICPR program.

Human pose recognition

TU Delft human torso detector

Customer story

At Delft University of Technology Prof. Emile Hendriks and his PhD and master students Fei Fei and Javier have developed a human-computer interface based on body pose recognition. In live video stream, they detect body postures by matching human torso models. Pose classifiers, trained on model parameters in PRSD Studio, are then quickly exported for execution in a real-time posture recognition demo written in C++ and OpenCV.

We have used PRSD Studio for design and evaluation of a real-time pose classifier. PRSD Studio is a very valuable and flexible tool and made our life much easier. Also the support was very good. I can recommend it to every pattern recognition system designer. -- Emile Hendriks, Associate professor, TU Delft.

More application examples

Visualizing classifier decisions in N-D spaces

Visualization of classifier decision in N-D feature spaces.

Feature Highlight

Visualization of classifier decisions in the feature space help us to understand classifier behavior. The visualization is straightforward when our data has only two features. But what about multi-dimensional problems? PRSD Studio provides a solution: the classifier decisions are visualized with respect to a reference data sample. The decisions are shown on a 2D cross-section of the multi-dimensional space in the two features of interest. This tool allows us to analyze classifier decisions in multi-dimensional features spaces.

See the video in our blog