A Prominent Object Region Detection Based Approach for CBIR application

Image retrieval based on visual content of image has constantly been a promising research field in the area of information retrieval to retrieve most analogous images from an image database according to a visual query image provided by user. Since the retrieval efficiency of all CBIR technique depends on the mining of appropriate salient visual features from images, it is still challenging problem to extract suitable visually prominent features from an image and use them in CBIR applications. A new CBIR scheme has been projected in this paper by using an object detection approach to locate prominent object region from images and subsequently some statistical parameters are computed from the detected object region for formation of image feature vector in CBIR application. Initially, the object region is detected and cropped using Graph based visual saliency map and connected component approach respectively. Later, the cropped object region is decomposed into several non overlapping blocks and some statistical parameters have been computed from each block. The proposed technique is tested on a standard Corel image database to estimate retrieval performance of the proposed image retrieval scheme. The experimental outcomes evidently confirmed that the proposed CBIR method is outperforming in object based images.