Defect classification of TFT-LCD with bag of visual words approach

Abstract

A defect classification algorithm with bag of visual words approach for thin film transistor liquid crystal display (TFT-LCD) manufacturing is proposed in this paper. Color and SIFT features are introduced to describe defect region. Visual words vocabularies are learnt separated for each features. The two features are separately coded in bag of visual words and combined by multiple chi-square kernel SVM. Classifier performances with different parameters are compared in experiments. Finally a good enough classifier for identifying 5 classes of defect is achieved.