Fast defect detection for various types of surfaces using random forest with VOV features

Abstract

Defect detection on an object surface is one of the most important tasks of an automated visual inspection system. The most modern defect detection systems are required to operate in real-time and handle high-resolution images. One of main difficulties in system applications is that it cannot be used for general inspection of various types of surface without tuning the internal parameters. In this paper, we demonstrate how to solve the problem mentioned above by using simple variance profile values of pixel intensities and applying it to the random-forest-based machine learning algorithm. Variance of Variance (VOV) profiles are used to describe the texture of an object surface and to amplify the irregularity of intensity variations. The feature amplification property of the VOV method can be applied generally to various types of surface and defect. For effective learning and reduction of false detection, a defect-size insensitive approach and a hard sample retraining process are introduced. The experimental results demonstrate reliable defect detection for various surface types without changing parameters.