Image Retrieval using SURF Features

Abstract

Modern technology has made image capturing cameras ubiquitous and increased accessibility to the Internet and Internet based services and soaring popularity of social networking sites for sharing pictures has resulted in large amount of images being shared over the Internet. The availability of cheap storage has allowed individuals as well as Web companies to maintain and host large collections of images. The huge collections of images pose a challenge in terms of efficient retrieval of desired images to both users and Web companies. The traditional databases have been found unfit for processing and retrieving images as they are incapable of capturing the information within the images. Traditional databases were designed with fixed text based information structures in mind and can store image information like resolution, metadata and even images as binary files but they cannot do image processing, classification and matching of images on the basis of information stored in them. This limitation of traditional methods is remedied using methods that are able to classify and retrieve images based upon the content that is seen by the end user and not the miniscule data captured by image metadata. The thesis discusses the implementation of image retrieval method used using image content as the differentiator between images. Also, discussed is a reproducible method to measure the performance of the proposed approach as well as the results of the performance of the image retrieval method on a standard image dataset.