Vision-based Clustering Algorithm Research And Application

Clustering is the things in accordance with certain attributes, the things gathered into categories, so that the similarity between the class as small as possible, within the same category of the greatest possible. Clustering is an unsupervised learning process, the same classification it is the fundamental distinction: the need to know in advance that classification is based on the data features, and clustering is to find this data characteristics, therefore, in many applications, as a cluster analysis a data pretreatment process is further analysis and processing data base.Cluster analysis is an important human behavior. Cluster analysis algorithm depends on the type of data, the purpose and cluster applications. In today's rapidly developing data mining and exploration data analysis, cluster analysis technique has been widely used in pattern recognition and image processing, biological, psychological, computer vision and remote sensing, and other fields. In practical problems, the traditional technique of cluster analysis prevailing inadequacies of the main performance in the following areas: clustering results of the sensitivity and initialization parameters strong dependence; difficult issue of the validity of the definition Clustering and reasonable it is difficult to determine the number of clustering; direct physical explanation can be poor.In recent years, the development of neural physiology and anatomy of Computer Aided by several highly accurate calculation of the primary visual system model, namely the visual system modeling of the various parts of different levels. Scale space theory is one of them, it describes quantitatively connected by lateral retinal image caused by the fuzzy effect.Through visual principle and algorithm combining scale space, the visual system to the structure was significant assumptions and stability of assumptions, using scale space clustering algorithm, have different levels of effective clustering. Major work:1, clustering algorithm comparisonThere are many kinds of clustering algorithm,the application needs to be involved in data types, the purpose of clustering specific application requirements and to select the appropriate clustering algorithm. Clustering algorithms can be roughly divided into the following categories: partition method, hierarchical methods, the methods based on the density, grid-based methods and model-based methods. Then various clustering algorithms specific comparisons drawn between the performance of different methods on different.2, the assumption that the structure of visual systemBy introducing visual system, Weber's Law, the structure of the visual system was significant assumptions and stability assumptions for the next chapter scale space clustering algorithm to pave the way. The structure of the visual system was significant assumptions: those nerve cells caused more excitement than the structure of nerve cells that cause less exciting structure is more important. The structural stability of the visual system assumptions: those in the larger scale can be observed within the framework of the structure of objects smaller scale than those in the range of objects can be observed that the structure is more important.3, scale space clustering algorithmFirst introduced scale space concept, introduced retinal biological model, the visual front-end system scale-space model, the focus on scale space clustering algorithm. Scale Space principle: When the full scale parameters hours, each data point is a category, and when larger scale parameter gradually, gradually small category of data leveraging on the latest advances become a general category of data. This classification of the results generated by the composition of a tree, the node representing different scale clustering of categories, the father node category expressed by the son represented by the node class from leveraging on the latest advances. This clustering algorithm is a gradual clustering algorithm, which contains a series of data classification. 4, the effectiveness of the clustering problemClustering validity of cluster analysis is a more difficult issue, which involves the algorithm data structure and explain the significance, such as the number of data in the category, obtained by the algorithms of this type is true, what kind of more a series of issues such as effective. Through four's explanation, to judge the effectiveness of clustering5, illustratedThrough numerical simulation experiments to illustrate scale clustering algorithm can be effective at different levels of clustering.