On benchmarking non-blind deconvolution algorithms: A sample driven comparison of image de-blurring methods for automated visual inspection systems

Abstract

This paper discusses motion blur reduction in digital images as a pre-processing step for automated visual inspection (AVI) systems. It is described how impulse responses of prevalent inspection set-ups can be modelled for efficient image enhancement. Common criteria for deconvolution performance measurements are listed and the results of a competitive benchmark of 13 state-of-the-art non-blind deconvolution algorithms are presented. Covered topics are illustrated by the example of a real-world inspection system for automatic quality control in woven fabrics. To meet real-time requirements, the efficient implementation of two selected algorithms based on GPU hardware is presented.