The Research On Algorithms Of Relevance Feedback And Classification In Image Retrieval

With the rapid development of Internet and multimedia technology, the amount of digital image is increasing rapidly. How to retrieve necessary image from large amount image efficiently and quickly has become increasingly important. Therefore, research on image retrieval is attracting more and more attention. Content-based image retrieval was put forward in 1990s, and it performs retrieval according to the visual features of images. Content-based image retrieval improves the precision and efficiency of image retrieval, which provides it with a capacious application fields. As key techniques of image retrieval, relevance feedback and image classification have become the hotspot in content-based image retrieval fields.As an important mathematical tool for dealing with inexact, uncertainty or vague knowledge, the advantage of rough set theory is that it can obtain the essential rule of the problem from actual data without other information, which makes it has been widely applied in the field of image retrieval in recent years. Emphasizing on this character, the thesis makes deeply research on relevance feedback and image classification with rough set as theory basis and content-based image retrieval as research background.Relevance feedback has effectively solved the problem of semantic gap in image retrieval and improved the system retrieval performance greatly. By systematically summarizing and analyzing existing relevance feedback algorithms, this thesis presents a method based on rough set. This method reduces the image features with the relevance knowledge of entropy, and the compute of comparability matching, which increases the image retrieval efficiency as a result. By adjusting the feature-weighting, the method emphasizes the interesting feature to retrieve again, which enhances the retrieval precision. The results of the experiments show the effectiveness of the algorithms.Image classification is the basis of image retrieval, and effective classification method can achieve the rapid image retrieval. The thesis inducts membership function into the image classification. With membership function, we can calculate the membership value of attribute set, and attain all the decision rules on decision table and the minimal rule set of reduced condition attribute set without core-valued. Compared with conventional Bayesian method, the results of the experiments in this thesis indicate the excellent performance in image classification field.The thesis achieves a content-based image retrieval system. The system provides an experiment tool for algorithms test and further research.