Mobile visual search system

• Main Author: Soh, Yu Hao
• Other Authors: Yap, Kim Hui
• Format: Final Year Project
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/61457
• Institution: Nanyang Technological University

The usage of mobile devices with an in-built camera is very common in today’s world due to the ease of usage and portability. Users can depend on their mobile devices to retrieve information easily with the help of the 3G network. This project helps users to identify coins from various countries by capturing the images of the unknown coins. In this project, the accuracy and robustness of various key point detection methods were evaluated. Coin database consisting of coins from 29 countries was created for this project. The images in reference database served as images stored in the server. Images in test database served as images taken by the user using mobile devices. Two key point detection methods such as Scale Invariant Feature Transform (SIFT) and Harris Laplace were proposed to improve system accuracy. Harris Laplace proved to perform better in system accuracy on coins. The accuracy of coins on system was further tested by comparing Harris Laplace detector incorporating Geometric Verification (GV) and without GV. From the results obtained, the incorporation of GV provided better accuracy. Lastly, the effect of multiple reference images as compared to single reference images in reference database on recognition accuracy was also studied.