With the rapid development of image processing and integrated circuits, more and more attention has been paid on vision inspection, which is one of the most important non-destructive detection tools. Researchers have developed a large number of techniques based on image processing for different detection purpose. Among which, clustering analysis plays an important role. Several new clustering techniques and their applications in vision inspection will be studied in this dissertation. The main contents involve: swarm intelligence clustering, kernel density estimation clustering, texture surface detection based on wavelet transform and co-occurrence matrix and X-ray fish bone detection.Feature extraction is an important task in clustering analysis and vision inspection, which influences the inspection efficiency and effectivity directly and greatly. Therefore, we firstly generalize the defect detection features from the four aspects such as statistical features, geometrical shape features, color features and texture features and their extraction and selection approaches.Based on the above, the contributions of our work mainly focus on the following aspects:1. One clustering technique based on ant colony algorithm is proposed. Ant colony algorithm, which is widely used for optimization tasks, is used for fuzzy clustering. According to its defect of large computation quantity when dealing with vast image data, two improvements are made: one is to improve the heuristic function to enhance the positive feed back ability, the other is to set initial clustering centers. By the latter approach, the comparison and computation among ants are changed to that of ants with clustering centers, which greatly cut down the computation quantities.Particle swarm in image clustering is tested. A fitness function is designed based on the principle of minimum-in-cluster-distance and maximum-between-cluster-distance. And particle swarm optimization, which is combined with fuzzy clustering, is used in image clustering.2. One optimal bandwidth kernel density estimation clustering approach is presented. Since the influence of bandwidth selection of kernel function is great on the result of density estimation, which is even larger than the selection of kernel function, an approach for bandwidth selection based on density estimation entropy is proposed. Then combined with mean shift, which is an efficient method for module searching, the bandwidth selection approach is tested for... |