Need for accuracy verification of machine vision algorithms and systems

Abstract

The author summarizes important parameters and procedures for verifying the measurement and recognition (classification) accuracy of machine-vision algorithms/systems. He asserts that current practice in this area is very inadequate, and proposes some measures to improve this situation. A survey of published machine vision research shows that the verification of algorithms/systems performance is very seldom done in an adequate manner, with claims made on the basis of 2-3 processed images.