Mobile visual search (MVS) has drawn great attentions in both the industry and research area. The MVS system will retrieve relevant ​information and display them on the mobile devices by simply taking a snapshot of an object-of-interest with a built-in camera. Visual search is still a challenging task in the robustness, computational complexity and scalability. Compared to the PC-based image retrieval problems, mobile visual search is more challenging due to the insufficient computing power and battery capacity and the constrained bandwidth of the wireless network.

The MSV research group led by Prof. Yap has built an efficient and effective MVS system to alleviate the problems in the Landmark recognition task. Non-linear local detectors are proposed to mitigate the problems caused by the cluttered background and the intensity changes. A new saliency-integrated mobile visual search framework is developed. The discriminative power of the proposed MVS system is enhanced by incorporating saliency information into the local descriptor extraction, visual codebook construction and image representation. Furthermore, saliency information is used to guide the sampling process in geometric verification to improve efficiency. Fast geometric verification algorithm is developed to reduce the computational complexity without the sacrifice of the system performance.