KIZUKI Processing for Visual Inspection: A Smart Pattern Pop-Out Algorithm Based on Human Visual Architecture

There are many cases in which it is difficult to detect defects by visual inspection. These defects, such as very small cracks and thin stains, etc., are slight difference and hide in background texture. Even if a skilled inspector can detect these defects, they are undetectable by basic image processing algorithms. This special ability of the inspectors is called "KIZUKI" in Japanese. This paper describes how to develop irregular region extraction programs based on human visual architectures. Two theories have been used to model the "defects pop-out = KIZUKI" process of visual inspection: the peripheral vision and the involuntary eye micro-vibration theories in the field of psychology of vision. We have implemented this model experimentally. It consists of two steps: forming low resolution images by variable re-sampling, and then detecting relatively salience regions at each low resolution image and integrating these regions. Experiment results indicate that our approach have been reasonable. Since our programs have been very simple and effective, it might be incorporable into different visual examination systems widely.