Framework for Improving Annotation-Based Image Retrieval Performance

As the proliferation of available and useful images on the web grows, novel methods and effective techniques are needed to retrieve these images in an efficient manner. Currently major commercial search engines utilize a process known as Annotation Based Image Retrieval (ABIR) to execute search requests focused on image retrieval. The ABIR technique primarily relies on the textual information associated with an image to complete the search and retrieval process. Using the game of cricket as the domain, we describe a benchmarking study that evaluates the effectiveness of three popular search engines in executing image-based searches. Second, we present details of an empirical study aimed at quantifying the impact of inter-human variability of the annotations on the effectiveness of search engines. Both these efforts are aimed at better understanding the challenges with image search and retrieval methods that purely rely on ad hoc annotations provided by the humans. Finally, we propose a framework that utilizes generic templates to aid the human's cognitive capabilities to fill relevant for annotation needed in a specific domain in a more systematic way. The systematic annotation will not only reduce the mental task load on the human, but also would increase the precision and recall of a search engine.