Efficient object identification and multiple regions of interest using CBIR based on relative locations and matching regions

Nowadays information retrieval systems get more attention due to the increasing use of multimedia technologies. The information may be in the form of video, image, sound and/or text. Application of surveillance, digital libraries, web applications and various other applications that handle enormous volume of data essentially have information retrieval components. This paper demonstrates an image retrieval system based on multiple regions that give a client interface for helping to identify the watershed regions-of-interest inside of an input image. The relationship between semantic ideas and visual elements is established by supervised Bayesian learning from positive bags. For comparison, feature vectors of regions which have similar regions code to the regions of query image can be used during retrieval. On standard datasets the proposed algorithm has been applied and accomplishes great annotation performance.