Login Register

Auto-login on future visits

Forgot your password?

Newsletter November-December 2013

Learn from local image information


How to compute local image features

How do you represent image data for classification? What image characteristics are relevant in your application? Very often it is local texture or appearance. How do we practically leverage the spatial connectivity of the pixels, and learn from local image information? In this tutorial we focus on local image features. We discuss how to extract them and illustrate why it is important.
Watch this 6 minute video to find out more.

Impressions from October 2013 perClass course

Group picture participants of the machine learning course for industry.

From the blog

In our October course, we welcomed participants coming from Israel, Nigeria, UK, Poland and The Netherlands, again a very multi-cultural group. This time we had a strong presence of academic participants. They are challenged in their research projects to develop medical diagnostic applications, quality control systems and image-based emotion classification. As we have learned, their shared motivation to join the course was the interest to go beyond theory and leverage machine learning as a working tool.
As one participant wrote: "I feel that I have returned to work as a stronger and more able developer, in better position to handle tougher and more challenging problems. I have no doubt participating in the course has strengthen my professional ability, and I strongly recommend to every PR algorithm developer out there to take part in it."

Spring 2014 course dates are being fixed: Tell us your preference!

fast Gaussian mixture classifier in Matlab

Ever wished for a faster Gaussian mixture classifier?

From the blog

Oh yes, we did! In many of our projects, the best classifier is something a bit more flexible than a simple model but not much more.The Gaussian mixture is an excellent candidate.
However, existing Matlab-based mixture implementations are simply too restricted in terms of training set size and training speed. That's why in perClass 4.1 we introduce a new C-based implementation of sdmixture that scales to large data sets. Read more

perClass new release


the fastest Gaussian mixture classifier in Matlab

perClass 4.1 is out. These are the main improvements:

  • New fast mixture of Gaussian classifier
  • Convolution-based image features
  • Crisp combination of classifier decisions
Much more in the release notes.
If you are a customer you can find the new release in the Customer Center.
Not yet a customer? Try out the demo!