AN AI VISION SYSTEM FOR THE INSPECTION OF COMPLEX COLORED OBJECTS

Many industrial inspection tasks can be performed satisfactorily by co nventional image processing systems which utilise feature extraction a nd related feature measuring algorithms. However, for the inspection o f parameters such as texture or colour, it also may be necessary to ad opt an AI technique which involves the use of artificial neural networ ks. This paper outlines one such system (WISE) which incorporates algo rithms for both conventional image processing used for position and ro tation normalisation, and n-tuple processing for subsequent pattern cl assification. Since its initial implementation, WISE has undergone con siderable development and is now capable of inspecting various types o f pattern space such as 2-D shape space and 3-D colour space. Results obtained from the use of these pattern spaces are documented as are re sults for 'travelling window inspection', 'zoned H-S-I inspection', 'r etinal simulation', '% inspection coverage', 'inspection sensitivity' and 'archetypal inspection artefacts'. In conclusion, the present high ly evolved WISE system provides excellent recognition performance for artefacts which are both complex and coloured. Although, at present, W ISE is implemented in software (C++) and, therefore, slow in operation , it is considered that, by using special-purpose hardware and/or para llel processing techniques, operational speed approaching TV frame rat e will be feasible in the future.