A multi-scale learning approach for landmark recognition using mobile devices

The growing usage of mobile camera phones has led to proliferation of many mobile applications. Landmark recognition is one of the mobile applications that are gaining more attention in recent years. The main idea of the application is that a user will use a camera phone to capture the image of a landmark or building and then the system will analyze, identify, and inform the user the name of the captured landmark together with its related information. A new mobile landmark recognition method is proposed in this paper: first, a set of multi-scale patches are extracted from the landmark images. Discriminative patches of the images are then selected based on a Gaussian mixture model (GMM). A combination of color, texture and scale-invariant feature transform (SIFT) descriptors are then extracted from the selected patches. They are used to train support vector machine (SVM) classifiers for each category of landmark. Experimental results using a database of 4000 landmark images illustrate the effectiveness of the proposed method.