Extracting information from a query image, for content based image retrieval

This paper describes a method to solve the problem of Object-Centric Content Based Image Retrieval (CBIR), motivated by concepts from theory of cognitive sciences. According to cognitive models, there are two lobes in human brain, one is responsible to solve the problem of object recognition, while the other solves the problem of localization (or detection). It is the exchange of mutual information (back and forth) between these two lobes which enables the human brain to simultaneously detect and recognize the objects in a complex scene. Based on the above cognitive paradigm, we propose a RADAR (Retrieval After Detection and Recognition) framework, to solve the problem of object-centric CBIR. Our proposed RADAR framework helps to simultaneously detect and recognize the object in a given test sample for retrieval task. Object-Centric CBIR retrieves a set of samples from a database of identified categories, similar to the categories in a query image using a matching criteria based on features extracted from the localized region(s). Results are shown using various real-world datasets (including PASCAL VOC) to exhibit the performance of the proposed method.