Implementation of Open-Source Image Analysis Techniques in Commercial Quality Control Systems

Abstract

Using image pattern matching techniques, this paper presents the development of an open-source based quality control program that can distinguish between sample pages in a book printing/production environment, intended to enable automatic rejection of incorrect pages from the binding process. Novel adaptations are made to an ORB (Oriented FAST and Rotated Brief) based image matching system that deals with quantifying the confidence of a match to ensure that even identical pages in the incorrect orientation are rejected during book binding operations. The program is subjected to a variety of quantitative tests to evaluate its performance and from these tests the effects of various parameters used in the program are discovered, allowing tuning to be performed. Potential paths for development of the analysis program are discussed, with the implementation of machine learning, highlighted as a possibility, to provide automated parameter tuning, and suggestions for further optimisation of the software are made, with a view to creating an even more bespoke version that retains performance whilst cutting computational overhead to a minimum.