EVALUATION OF FEATURES FOR DEFECT IMAGE RETRIEVAL

摘要

In this paper different feature descriptors are evaluated for content-based retrieval of defect images. Especially the visual descriptors of the MPEG-7 standard are of interest. The evaluation is done with a small, pre-classified defect image database using the K-nearest neighbor (KNN) classification. The best features are then applied to a content-based image retrieval (CBIR) system called PicSOM. The results indicate that some of the MPEG-7 descriptors work very well with our defect images, but there is a need for some additional shape descriptors.