Effective inspector for detecting foreign substances in bottles with inhomogeneous structures

In order to solve the problem of high costs and low efficiency caused by manual inspection, an automatic inspector for foreign substances in bottles with inhomogeneous structures based on machine vision technology is proposed in this paper. First, we extract the region of interest based on meanshift segmentation and align the images by registration and rectification. Then an adaptive image variation detection method is established to locate the potential foreign substances. To avoid the brightness disturbances caused by inhomogeneous structures on the bottles, an occurrence probability image which models the probability of each changed pixel to be true foreign substance is learned and candidate foreign substances are obtained by taking account of both the probability distribution and brightness variation. Finally, SVM classifier is applied to further identifying foreign substances based on their appearance features. Experiments show that this inspection algorithm has satisfactory detection accuracy and can greatly inhibit false detection caused by inhomogeneous structures.