Robust Printing Defect Detection on 3D Textured Surfaces by Multiple Paired Pixel Consistency of Orientation Codes

Abstract

A new method for defect detection in 3D textured (SHIBO) surface is proposed. In the presence of 3D textured surfaces, and/or illumination change such as shadowing or highlighting, many conventional ways for visual inspection have been proved as of some troubles. For such purposes, it is proposed to utilize the orientation code based consistence of multiple pixel pairs to detect defects on character which is printed on 3D textured surface. This algorithm consists of two stages, they are training stage and detectingstage. Training stage is for making defect-free model based on multiple paired pixel consistency. Detecting stage is to identify whether the target pixel match its model. Experimental results for real products image demonstrate the effectiveness of the proposed method for defect inspection and location.