4 - Classification géométrique par polytopes de contraintes. Performances et intégration

We present in this paper a fast classification operator suitable for image processing, the performances of this operator as well as its implementation in the form of an ASIC. In image segmentation and classification in view of defect detection, it is often impossible to find a reduced set of pertinent characteristic parameters which allows to distinguish the classes. We propose herein a geometric classification method by stress polytop training which allows the use of a great number of parameters and ensures a high decision speed . The decision operator associated with the classification has been implemented in Standard Cell and Full Custom . Its ease of use, rapidity, and robustness in classification are the major qualities which enable it to compete with neural operators .