Image Retrieval Based On Formal Concept Analysis

With the development of the science technology, the object and scale which the modern information retrieval deals with have a huge change. Therefore, people need a kind of technique urgently that can seek an interview image quickly and accurately. This is the image retrieval technique. The process of image retrieval reflects the relationship of concepts which are formed by the feature extracted from the image data.Meanwhile, the Formal Concept Analysis (FCA) theory provides one kind of model to organize the data by the concept lattice. The nodes of concept lattice embody the unification of concept content and extent, suited very much to discover knowledge. Therefore, FCA is acquired by an extensive application in many realms, like information index, the numerical library, software engineering, knowledge discover and etc..In this thesis, domestic and international image retrieval status and the application of FCA in relative area are introduced, and then a novel method to retrieval image according to the formal concept analysis is proposed. The main content includes:First, the existing method of extract features which have good discretion abilities with different semantic is introduced. Through the analysis of the information to the picture and existing low-level feature, these methods can overcoming to apply the same feature extract algorithm to calculate all image feature's limit before.Next, based on the membership function of fuzzy mathematics, a set of candidate classes for the query is extracted. This can reduce the index scope consumedly for have no need to compare all images in database.Then, based-on FCA, each query is represented by an individual concept lattice instead to construct all images in one concept lattice and therefore the complication can be reduced. Through concept match to decide how similar the nodes in a query concept lattice are to the nodes in an image concept lattice of candidate classes. Output the images ordered by the similarity degree.Finally, the results of image classification by using FCA-based and Color histogram-based methods are compared in Coral image database. The proposed method can improve retrieval efficiency and reach a higher retrieval precision. The time needed for returning retrieval results and effectiveness of our method is demonstrated.