The use of convolution operators for detecting contaminants in food images

摘要

This paper describes a monitoring system that detects contaminants (such as metal, glass, stone, wood, rubber and plastic) in food stuffs. The X-ray images of the food stuffs are highly textured. The contaminants do not necessarily exhibit a separate texture property to the food substrate, hence we address the task of detecting defects or anomalies within the texture of the food background. No a priori knowledge of the type of contaminants is assumed. The methods used are based on convolution filtering, where the convolution masks act as matched filters for certain types of textural variation found in the images. Three different techniques have been compared and contrasted and they include: masks with pre-set coefficients, masks whose coefficients have been generated using statistical methods and artificial neural network processing.