Film Defect Inspection Features and Classification

• Main Authors: Nai-Ting Chang, 張乃婷
• Other Authors: Chiou-Shann Fuh
• Format: Others
• Language: en_US
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/01840423354319449288

碩士 === 國立臺灣大學 === 資訊工程學研究所 === 100 === 　　We research on automatic defect inspection and classification on film. The five main defects on film are black spot defect, white spot defect, crease, white with black defect, and comet defect. We use automatic image processing to derive film defects and calculate 12 features, including geometric features, intensity features, and histogram-based texture features. We research an automatic best feature selection step for computing spread factor of each feature and each class. High spread factor means feature concentrated in respective classes and widely spread between classes. Find the best features with the highest spread factor in each class. After the inspection and feature selection, use Adaboost and back-propagation neural network to classify all defects.