A real-time automatic defect classification scheme is proposed to understand the situation of color-filter production procedure by analyzing the defect types. The proposed system consists of three stages: defect extraction, feature description, and defect-type classification. A reflex-lighted image and the same-area back-lighted image are both used for the defect classification. Four features extracted from the reflex-lighted images describe the shape of the defects and other four features extracted from the back-lighted images describe the appearance of the defects on the color-filter cells. A neural-network decision tree classifier is then designed for classifying the defect types. The neural network has the ability to analyze the complicated non-linear relationship between input signals and the desired outputs. A back propagation algorithm is used as the training method for the proposed classifier. The experimental results show that the proposed scheme can efficiently classify the defect types and the proposed neural-network classifier is superior to other classifiers.
INTERNATIONAL JOURNAL OF INNOVATIVE COMPUTING INFORMATION AND CONTROL