단일 결함 영상을 사용하는 자동 결함영역 분리 방법

초록

In electronics and semiconductor industries, defect classification plays an important role in identifying the cause of defect occurrence. To increase the performance of defect classification, unnecessary information such as background needs to be eliminated due to the possibility of influencing classification result. I n this paper, we propose an auto isolation method which does not require any reference image to crop the defect region only. This method uses similarity between sub-images and modified converging squares algorithm to extract the defect region. Similarity between sub-images is used to visualize the defect region even in multi-background defect image. Converging squares algorithm is specifically modified so that it may locate the defect region even in noisy similarity result. The method was able to correctly crop the defect area in 93.94% of mono-background defect image test set and 79.94% in multi-background defect image test set.