Location-aware mobile information hub

Abstract

In recent years, 3G network has gain popularity which leads to increased consumerism in the area of smartphones. Smartphones allows the user to retrieve information anywhere as long it is connected to its subscriber. This promotes the habit for mobile user to retrieve information immediately when in doubt of something. This project aims to help mobile users retrieve landmark information through the capturing of images. Landmark image recognition has been around for some time but its development in the context of Singapore has been limited. In this project we will study the reasons of the image that will constitute to the overall accuracy of the system and the limitations of the database such as undesirable features. Undesirable features are unwanted features that are extracted from the image which deranges the focus of the image. Geometric Verification (GV) and saliency detection have been proposed to tackle these undesirable features. GV shows a slight improvement in terms of accuracy (1-2%) with a downside of large increase in recognition timing. Saliency detection on the other hand, presents a significant improvement in terms of accuracy (> 7%) with a compromising slight increase in the recognition time (~0.1s). Lastly, we study the impact of different image size used in feature extraction for evaluation of the system performance. Moving forward, we can explore on incorporating GPS information stored in the images to narrow the database for matching, thus increasing the speed of the system.