Machine vision for crack inspection of biscuits featuring pyramid detection scheme

Abstract

One of the challenges associated with machine vision inspection of biscuits or baked products with non-uniform colour distributions and textured background is the detection of a small and minute crack. In this study, a pyramid automatic crack detection scheme was proposed. This requires an enhancement method to properly distinguish the crack and intact samples. Canny–Deriche filter was used to emphasis the crack and reduce the noise. In order to segment minute crack pattern with less noise, a unimodal thresholding technique was developed and tested. The detection was based on support vector machine (SVM) featuring Wilk’s λ selection criteria. The accuracy of the system was compared with standard discriminant analysis. It was discovered that the pyramid SVM after Wilk’s λ analysis was more precise in detection compared to other classifiers, resulting in the specificity and sensitivity of 98% and 96% respectively, and average correct classification of consistently more than 97%.