5 
  
329 
  
Fig. 6 The VME-Bus Colour Brain processor performs real-time LUT classification of multisensorial 
camera signals having up to a 21 bit-wide feature vectors per pixel 
Pixel-wise LUT classification reduces the [ n*m] *k "vectorial" image into M scalar [ n*m] binary images, n,m 
being the number of columns and lines of the image, k the dimension of the feature vector and M the number of 
trained classes. There is little loss of information because a binary class label image retains all the important 
information on objects in a scene ( class membership as an identifier, geometric information etc. ). 
Fig.6 shows our VME-Bus Colour Brain processor, which contains all the electronics to accept 24 bit of 
digitized video signals, perform hardware shading correction and 21 bit hardware LUT classification. It is used 
for real-time colour classification but can as well perform classification of multisensorial camera signals up to a 
maximum of 21 bits per feature vector. 
S. Applications of a Colour&3D camera 
We are currently working within the of the German BMFT project , Electronic Eye - Optical Recycling on a 
system for sorting plastic waste by colour, composition, dimension and density. The combined imaging sensors 
are colour CCD line scan cameras , a 3D line camera, an imaging NIR sensor array and an X-ray line camera. 
Multivectorial classification is one of the non-hierarchical classification and data-reduction schemes we are 
working on. Other interesting industrial applications especially for the Colour&3D camera are the quality 
inspection of wood panels, of food , of bakery products, of granular products etc. We will report on these in a 
near future. 
Acknowledgement 
We are grateful for the financial support of the German Federal Ministery for Research and Technology for 
some of the research aspects covered in this paper through funding within the Electronic Eye“ project. 
References 
/M Massen, R.et al. Colour and Shape Classification with competing paradigms: Neural Networks versus 
Trainable Table Classifiers. ECO 3 Int. Congress Optic. Science, Den Haag, 1990 
/2/ | Massen, R. et al. Real-time Colour Classification for Preprocessing Photogrammetry Images. ISPRS 
Symposium on Close-Range Photogrammetry and Machine Vision, Zürich, 1990 
/3/ Massen, R. Int. PCT patent application PCT/EP94/03621 
/A| | Seitz, P. Annual report of the Paul Scherer Institute, Zürich, 1992 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop “From Pixels to Sequences’, Zurich, March 22-24 1995