Kouretes Colour Classifier
The proposed approach is based on labelling by hand a representa- tive set of images from the robot camera and training a classi er which generalises over the entire color space. The illumination problem is addressed either by including special illuminant features (average color values from a large region or from the entire image) in the classi er, or by transforming the input to an appropriate reference illumination level through histogram speci cation before classi cation. These procedures have been integrated into the Kouretes Color Classi er (KC 2) graphical tool, which provides intuitive means for labelling images (regions, clusters), selecting features (neighbourhood, illuminant), training classi ers (Decision Trees, Support Vector Machines, Neural Networks), and generating code for ecient execution on the robot. The tool minimises the user time required and delivers excellent color classi ers using only a few images.
Me concernant
o Computer Vision (Computational Geometry, Differential Geometry, Surface Analysis) o Machine Vision (Colour and Landmark Recognition, Feature Extraction, Image Enhancement) o Machine Learning (Decision Trees, Neural Networks, Self Organising Maps, Clustering) o Quality Assurance (Automated Visual Inspection (AVI), 3D Solder Paste/Joint Inspection) My research interests involves the characterisation of solder paste in SMT process using 3D shape acquisition technologies. At first, we need investigate the correlation between solder joint defects and solder paste volume and shape deformities, using afterwards that correlation to create a model of stencil printing process for solder paste shape and volume optimisation.
$ 30 $ US / h
NOUVEAU FREELANCE !
