Vision 2011 trade fair
Last 8-10 November we have presented our perClass software suite at the Vision 2011 trade fair in Stuttgart. This is one of the largest European fair in machine vision, with exhibitors ranging from machine vision components, to complete system solutions, from software to services.
Companies and visitors came from across the globe. We really enjoyed meeting many people interested in adding more intelligence to their machine vision systems or improving performances. The general feeling is that people expect automation to deliver more. Moreover, dealing with natural objects greatly increases the challenges of automatic classification systems. People are looking for new or better solutions. Our visitors were intrigued by the possibilities of statistical machine learning in perClass. With perClass they can quickly develop their own solutions in-house, saving lot of time both in the design phase and when embedding the classifier in their application. See our new brochure.
And, by the way, thanks to all of you that came by to our stand. It was very nice to see you all in person!
APR course October 2011
The Advanced Pattern Recognition course has been a nice experience of meeting participants coming not only from different european countries but also from South Africa. Their applications fields are very broad, ranging from medical imaging to behavior analysis, defect detection and monitoring. They enjoyed the course and challenged each other in the final competition.
How to visualize errors in scatter plot?
Last time, we discussed how to identify samples that fall in a specific entry of a confusion matrix. Today, I will show you how to visualize them in the interactive scatter plot. This allows us to investigate the errors closely and quickly understand which feature space region is most affected.
(0) Comments
How to find out what samples are misclassified?
Often, we need to get more detailed understanding of our classifier errors. Let us go through a simple step-by-step example how to retrieve misclassified examples.
Read more...(0) Comments
Sponsor at the Brain-computer Interface workshop
Brain-computer interfaces (BCIs) use brain signals to communicate user’s intent. This allows a person with severe disabilities to express herself just by thinking. Neuroscientists, clinicians and biomedical engineers gathered together for the 8th BCI symposium organized by Nick Ramsey’s group at the Utrecht Medical Center. 
PR Sys Design sponsored the workshop on 18-19 May. In our talk we have illustrated important pattern recognition concepts, such as different classifiers options, tuning of a classifiers, and building of detectors and classifiers cascades. Most importantly, we have shown that all this can be immediately integrated in the BCI2000 platform using PRSD Studio Runtime library. The BCI2000 is a general-purpose system for brain-computer interface (BCI) research. It can also be used for data acquisition, stimulus presentation, and brain monitoring applications. It has been provided to about 600 laboratories around the world. Thanks to Dr Erik Aarnoutse and Dr Aldemar Torres Valderrama there is a module for linking PRSD Studio classifiers directly into the BCI environment. PRSD Studio classifiers can be run on live data. In a two-hours session we have assisted the researchers to walk through all pattern recognition steps, from training and tuning of classifiers to their embedding in the BCI2000 platform. Participants enjoyed exploring the power of pattern recognition with practical hands-on exercises on real brain signals data. We are very happy to support and provide tools for this research which has such a high impact in the life of disabled people.
Tutorial at MISP 2011
We have been invited to offer a tutorial on pattern recognition at the Medical Imaging Symposium for PhD students. On May 6 we joined the MISP meeting with many PhD students coming from all over The Netherlands. In our talk we outlined pattern recognition techniques typically used in medical imaging and pointed to specific problems arising in medical applications. The tutorial was structured in two parts. Firstly, we discussed statistical pattern recognition methodology and surveyed number of basic concepts and techniques. In the second part, we highlighted number of challenging issues that are typically appearing in medical imaging applications, illustrating them with practical examples. The atmosphere and the discussions were stimulating and joyful. It has been a pleasure to meet all these bright researchers. We wish them success and fun in their interesting medical projects!
Demo at WIAMIS 2011
On 13-15 April TU Delft organized the 12th International Workshop on Image Analysis for Multimedia Interactive Services. WIAMIS is one of the main international fora for the presentation and discussion of the latest technological advances in interactive multimedia services. We gave a talk on the challenges in classifier design and deployment in computer vision applications. We also presented a joined demo with Prof. Emile Hendriks and PhD student Feifei Huo of the Vision Lab at TU Delft. 
The demo is a pose-driven game prototype. The live video from a camera is analyzed, the human torso is detected and the classifier identifies the person’s pose. The real-time classification part of the game is performed with PRSD Studio. The trained classifier that has been trained in Matlab, runs directly in the C++ interface made by Feifei. The game responds to a set of predefined poses. We have demonstrated change of a real-time classifier in a running application without stopping or recompilation step. This allows us to quickly test impact of pattern recognition techniques such as rejection in a live game. Many people were interested and came along to play. A nice way to combine scientific research with fun!
APR course April 2011
Last week we have been teaching in the Advanced Pattern Recognition course. Each participant came from a different country. This time all participants were from Europe, as one person from Sudan had to cancel last minute due to a visa problem. Although all participants had already years of experience working in pattern recognition, their feedback on the course has been very positive. They were happy to extend the range of methods they had used so far with new ones, and enjoyed the broad spectrum of topics addressed in the course. We wish them further successful pattern recognition projects.
(0) CommentsEasy class subset with regular expressions
Quickly selecting subsets of classes helps you to inspect and learn from your data.
PRSD Studio now introduces a new, simpler way of selecting classes using regular expressions. Regular expression is a string describing a pattern matched to the class list. The simplest use is to select classes using a substring. In this video you will learn how to use regular expressions directly in the scatterplot, or while relabeling the data or creating a subset from the command line.
This video requires a more recent version of the Adobe Flash Player to display. Please update your version of the Adobe Flash Player.
(1) Comments
Tutorial on how to build accurate and fast k-NN classifiers
k-NN classifier is often adopted for complicated problems with high overlap. Although it may provide high accuracy solutions, it suffers from slow execution. In this video we show how to limit the set of k-NN prototypes making it both accurate and fast in execution.
This video requires a more recent version of the Adobe Flash Player to display. Please update your version of the Adobe Flash Player.
(0) Comments