Multi level object relational similarity based image mining for improved image search using semantic ontology

Abstract

The problem of image mining has been well studied in literatures and there exist number of approaches towards this issue. But, the most approaches consider only the spatial features rather than the meaning of the images. This deviates the modern image search from the search engines, because the users look for more meaningful results from the search engines. To support the search engines to improve their results, a multi level object relational similarity measure based image retrieval algorithm is presented in this paper. The method works in two levels, first the manifolds are trained by extracting the objects and assisting the labeling with supervised learning process. For each class of images, multi level semantics are maintained which consists of object features and their semantic meanings. At the test phase, the method computes multi level object relational similarity (MORS) measure for each semantic class. Based on the MORS value, a single semantic class has been identified. Based on identified semantic class, the result has been populated to the user. The proposed method improves the performance of image mining and improves the performance of image retrieval.