iGAPSearch: Using phone cameras to search around the world

This paper proposes and implement an iGAPSearch System (GPS Aided Photo Search System) to identify buildings through their photos captured by phone cameras. User need to take a picture of the building he/she wants to know with Android phone and upload the picture to our system. The system returns name and introduction on the buildings. We use SIFT features and the bag of features method to represent and recognize building images. Four distance methods are compared to estimate the distance between query image and those in database, namely L1-norm, L2-norm, KL-divergence and χ2 distance. We calculated ROC curves to compare these methods. Experimental results exhibit that χ2 distance has the best performance. Our system is fast enough for realtime application and its accuracy is satisfactory under the experimental environment.