Mobile visual search

Abstract

Nowadays, along with the development of mobile industry and the availability of network Bandwidth, mobile visual search (MVS) engine becomes a more and more fascinating research field. The concept of mobile visual search can be explained in the form of a fundamental pipeline, which includes three basic processes in sequential order: image feature extraction, local features matching and geometric verification (GV).

This project focused on one part of MSV which is local descriptors matching. Query image is extracted by detector for its local descriptors and these local descriptors are compared with local descriptors of database images to find matching image. One popular method of local feature matching is using vocabulary tree (SVT).

Implementation of SVT Bag of Word model using vocabulary tree and inverted file are to be conducted. Based on results of these experiments on the two algorithms, the project is able to evaluate the performances as well as efficiency of them. The experiment is also conducted with various set of image dataset to provide better estimation about the algorithm.