| After its first appearance in the 1980s as a simple device that receives telephone calls over a radio link, the mobile phone nowadays has become a powerful device with various features including mp3 player, high resolution camera, Bluetooth, Wi-fi, video/voice recorder, and so on. Such features have an immediate and extensive impact on our daily lives. The intense hardware-wise development of built-in cameras in mobile phones adds a entire new layer of functionality to the mobile device. These mobile hardware developments are changing the way people create, consume and communicate visual content. Among the broader uses of camera applications on mobile handsets, one of the most notable examples is mobile visual search.;In the near future, people would be able to enjoy the merits of wearable devices that are capable of taking photos, transmitting photos and displaying related information. No matter where people go, the wearable devices are able to provide information about the surroundings. One of the most important enablers of this technologies is large scale mobile visual search, where photos can be searched over a huge database.;Visual search over large image repositories in real time is one of the key challenges for applications such as mobile visual query by capture, augmented reality, bio-metrics based person identification, and effective management of billions of images being uploaded on the web. The search accuracy and response speed are two important performance factors. In this work, we focus on various aspects of this technology that enables large scale visual search.;The thesis covers the entire pipeline of large scale visual search. The major component of the thesis includes local feature compression, local feature selection, global descriptor hashing, the order statistics based feature matching, image re-ranking and multiple basis subspace selection. The technology developed in this thesis is used to develop a visual search prototype on Northwestern University campus. |
