Retrieval and ranking of biomedical images using boosted haar features

Retrieving similar images from large repository of heterogeneous biomedical images has been a difficult research task. In this paper, we develop a retrieval system that uses Haar features as its weak classifiers and builds strong training models using the adaboost algorithm. Our system is trained for each image category separately and the final boosted model is stored during the training phase. In the test phase, the most similar images for a given query image are computed using these boosted models. The main advantages of the proposed system are (1) cheap computation of the most relevant features for each image category and (2) fast retrieval of similar images for a given query image. Using performance metrics such as sensitivity and specificity, our results demonstrate the robustness and accuracy of the proposed system.