A Combination of Deep Learning and Segmentation Algorithms Applied to Appearance Inspection Problem

摘要

The separation of required object from given image is a critical issue toward the recognition and analysis phases in many image processing tasks. This paper describes an efficient algorithm used for detecting the defects and measuring their sizes in the appearance inspection problem. The suggested algorithm consists of the combination of deep learning and segmentation algorithms. The suggested method is applied to the appearance inspection problem and its detection rate of defects is 99.2% for 133 test images. The average error of defect size estimation by using the segmentation algorithm is 7.9% for 26 defects.