perClass applications

With perClass you can teach computer to automatically classify things. If you can collect examples of your objects of interest, typically images or signals together with annotations (what is it?), perClass can help you to build automatic system making such decisions on new observations.

See what others built with it. Let us know what is your challenge. We can help you to make the first steps!

Metal sorting using Hyperspectral imaging

hyperspectral metal sorting using machine learning and perClass Researchers of university Liège have uased perClass to sort scrape metal using hyperspectral imaging. The system uses two spectral ranges (VNIR and SWIR) to classify different types of metals such as copper, zinc or aluminum. perClass has been used to train spectral classifiers and to deploy them real-time in a robotic sorting system. Read More.

Detect accidents in a tunnel

real-time accident detection in a tunnel Joanneum Research has used perClass to develop AKUT system for acoustic detection of traffic accidents in a tunnel. The system responds to a range of sound patterns such as burst tire or braking sound 2 minutes faster than the current video-base surveillance solution (preventing up to 70 cars entering the danger zone in case of accident) Read more.

Classify food defects

defect classification of tomatoes with hyperspectral imaging Wageningen university (GreenVision) specialists detect food and plant defects with perClass classifiers trained on (hyper)-spectral images. The applications range from automatic harvesting to quality control. Classifiers are deployed into several industry-standard platforms such as HALCON, LabView and .Net. Read more about in-house training at WUR in Wageningen.

Sort rocks based on color and texture

rock classification by color and texture Rock sorting solution was developed classifying different mine-specific types of minerals based on multi-scale color and texture features. The final solution was validated to perform better than human geologist labeling identical rock images.

Recognize gender from face images

gender classification from a face image Gender classifier was built with perClass to discriminate males and females based on a single face image. The solution may be deployed with perClass Runtime both to PCs and to ARM-based systems such as Raspberry Pi.

See video of gender classifier in action

Detect thought patterns in brain signal

brain computer interface classification with perClass Group of prof.Nick Ramsey at Utrecht Medical Center uses perClass to classify mental states of paralyzed patients. Their Brain Computer Interface (BCI) serves as a platform for research of neurological processes See the publication in Journal of Clinical Psychology.

Segment bones in medical scans

medical image segmentation using extracted features and classifier Leiden University Medical Center (LUMC) researchers used perClass to develop a novel method for automatic segmentation of join replacement scans. Contrary to traditional, manually-tweaked image analysis algorithms, their solution learns to classify different anatomical structures using systematic pattern recognition approach. Read the publication in the International Journal of Computer Assisted Radiology and Surgery.

Segment cell structures in molecular biology

perClass used by Max Planck Institute In the Max Planck Institute of Biophysics cutting edge research is performed on neuronal organization in the mouse olfactory system. The researchers aim to describe the axonal wiring of olfactory sensory neurons in the living brain, using a combination of genetics approaches, in vivo imaging and behavioral studies on gene-targeted mouse strains.

"For analyzing our high content methods for profiling neuronal gene expression patterns in 3D we chose perClass Matlab Toolbox to aid in building sophisticated classifiers that find patterns unnoticeable to the human biologists. Marrying the latest in high throughput screening methods with custom analytical pipelines built with perClass Toolbox lets us ask questions previously outside grasp of molecular biologists." -- Dr. Bolek Zapiec

Classify prostate cancer in ultrasound images

Advanced Medical Diagnostics (AMD) has brought to market a new generation of cancer-detection ultrasound technology, providing physicians with clinically-meaningful visualization of prostate cancer.

"We use perClass to train our tissue characterization algorithm, that is to train a classifier to detect abnormalities in ultrasound data taken in clinical setting. We work with huge data sets and dozens of features, and perClass is able to handle them without performance issues or glitches. The interactive visualization tools provided by perClass allow us to explore our feature space and understand how separation could be achieved. We work often with the k-NN classifier and we enjoy the wealth of options to configure the classifier and experiment with various implementations of the k-NN. On top of perClass we use the consultancy and training services provided by PR Sys Design. They are always delivered in the most professional way and allowing us to expand significantly our knowledge in pattern recognition as well as in the perClass library itself." -- Dr.Dror Nir, CEO

Detect animal behaviour

Noldus is a market leader in behavioral research, providing solutions to study both human and animal behavior. Noldus is leveraging perClass for development of robust detectors and classifiers.

"We use perClass to build classifiers for rat behavior detection. I especially enjoy the metadata functionality and the interactive visualization where you can view subsets based on the metadata. It gives me insight in the data, and tells me what is feasible. Therefore, it saves me the time I would otherwise spend on useless efforts to optimize what is beyond reach. Also, I find the documentation and support really good!" -- Elsbeth van Dam, R&D Noldus

Update Jan 2014: Pattern recognition algorithms, investigated during our joined project, are now part of the EthoVision XT software.

Classify mineral particles by 3D shape information

DebTech, a company within the DeBeers mining group, developed their R-Sputnik particle characterization system using perClass. The system captures silhouette images of particles and generates 3D models of the particles.

"Using perClass we have implemented a classifier that takes in various features extracted from these 3D particle models and classifies them into a number of shape categories that are of relevance to us. The system also allows us to take simultaneous reflectance image of each particle. Relevant data from these images is fed into another classifier that enables us to obtain a colour classification for each particle." -- Dr. Garry Morrison, DebTech

Sort luggage at the airport

perClass helps Type22 to design robust classification algorithms and quickly embed them in their baggage handling solutions, minimizing the deployment time.

"With perClass we have been able to apply our classification algorithms to a variety of "real-life" baggage handling products. perClass development environment is easy-to-use, powerful and extensive. Deploying the created algorithms is simple and reliable." -- Jorick Naber, Type22 BV

Recognize human body pose in real-time

Get the Flash Player to see this player.

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. Note that this demonstrator does use a single standard camera, no Kinect or other depth sensors. Pose classifiers, trained on model parameters in perClass, are then quickly exported for execution in a real-time posture recognition demo written in C++ and OpenCV.

"We have used perClass for design and evaluation of a real-time pose classifier. perClass 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." -- Dr.Emile Hendriks, Associate professor, TU Delft.

Segment heart in 3D medical scans

Thomas Karavides, MSc student of Erasmus Medical Center in Rotterdam, The Netherlands, used perClass to optimize detectors for landmark points of 3D segmentation of heart. Thomas leveraged interactive perClass tools for ROC analysis to fine-tune his AdaBoost detectors according the specific performance requirements.