Content-based image retrieval based on fuzzy sets theory and learning automaton

This study presents a fully automated content-based image-retrieval method using the combination of fuzzy algorithms, fuzzy sets, and learning automaton. The major contribution of this study is to provide a method to automatically query image database with a high success rate of relevant images returns. This study tested for features and object class robustness on query image and on a database containing 200 colored images of different categories and 20 synthetic images that contain these objects: circle, square and diamond. The performance measures based on precision and recall were calculated while using various ranges of input parameters such as fuzzy membership functions, learning rate, and iterations, learning algorithms for learning automaton.; Content-based Image Retrieval (CBIR) involves retrieving images similar to an example query image in terms of some features extracted from the images. Uncertainty pervades every aspect of CBIR. This is because image content cannot be described and represented easily, user queries are ill-posed, the similarity measure to be used is not precisely defined, and relevance feedback given by the user is approximate. To address these issues, we proposed two parts method using fuzzy sets theory and learning automaton. The first part is the fuzzy sets theory: Fuzzy sets were used to model the vagueness that is usually present in the image content, image indexing, user query, and the similarity measure. This allows us to retrieve relevant images that might be missed by traditional approaches. The plethora (overabundance) of aggregation connectives in fuzzy set theory permits us to define a similarity measure that is tailored to the applications domain or the user's taste. The second part is the learning automaton for searching the database: Since the CBIR heavily relies on user-dependent weights (i.e. user profile), learning automaton offers an improvement of users profiles via the users' relevant feedback in searching for similar images in the database. Our method helps control the amount of time the user spent on marking relevant and irrelevant images and feedback to the system. This is done by the user changing the threshold values. A user can search for any image up to a threshold of 0.99.