A general purpose image inspecting system has been developed for automatic flaw detection in industrial applications. The system has a general purpose image understanding architecture that performs local feature extraction and supervised classification. Local features of an image are extracted from the compactly supported wavelet transform of the image. The features extracted from the wavelet transform provide local harmonic analysis and multi-resolution representation of the image. Image segmentation is achieved by classifying image pixels based on features extracted within a local area near each pixel. The supervised classifier used in the segmentation process is a fuzzy rule-based classifier which is established from the training data. The fuzzy rule base that is used to control the performance of the classifier is optimized by combining similar training data into the same rule. Therefore an optimization is achieved for the established rule base to provide the maximum amount of information with the minimum amount of rules. The experimental results show that the features extracted from the wavelet decomposition give contextual information for the test images. The optimized fuzzy rule-based classifier gives the best performance in both the training and the classification stages. Flaws in the test images are detected automatically by the computer.
展开▼