A graph based object description for information retrieval in digital image and video libraries

Search algorithms for objects of interest (OOIs) related to shape similarity in a video or image library have been implemented by various research groups. This work focuses on the search of a sample object (a car) in video sequences and images related to the shape similarity. We also investigate a new description for cars, using relational graphs. The goal of this study is to investigate the shape matching method based on a relational graph of objects, with respect to its accuracy, efficiency and scalability. The aim is to annotate the images where an OOI is present. Then, text-based querying can be performed to extract images of the OOI from a pre-processed database. The graph-based description of the object, with its meaningful parts, provides an efficient way to obtain high-level semantics from low-level features. The hierarchical segmentation increases the object detection accuracy in transformed and occluded images