Defect or Design? Leveraging the Angle of Opportunity for Detecting Scratches on Brushed Aluminium Surfaces

Abstract

Design features such as polishing strokes share similarities with defects; this makes defect detection and quality assessment difficult to perform both manually and automatically. Human assessors rotate objects to probe different incoming illumination angles and evaluate the defect dimension to limits samples i.e. decide whether differences between defect candidates and design features qualify as a defect. This process has poor access to quantifiable defect descriptors needed for automation and expose a gap in the existing evaluation of defects. To integrate this notion into automated defect detection we propose a spatio-temporal image acquisition setup capturing the defect descriptor Angle of Opportunity (AoO) which can be used as a feature for image-based classification. The Random Forest approach classified defects with an area under the ROC-curve of 92%.