Mobile product recognition for information retrieval

Abstract

Mobile applications are becoming increasingly popular as it can be easily downloaded and installed onto smartphone. It is also convenient to use as a smartphone itself is a portable device. Image recognition is one possible and useful application that can be developed. Even though the title of the project is Product Recognition, the actual project is on Landmark recognition. As the basis of the project is the same, hence the database used will not have significant impact on the project. In this project, the theory of the Bag-of-Words framework is covered and the performances of different sampling methods are evaluated though experiment. Scale-Invariant Feature Transform (SIFT) is chosen as the descriptor, with the Difference-of-Gaussian function as the detector for keypoint sampling. In summary, the dense sampling performs better than keypoint sampling.