Visual product search system for mobile commerce

• Main Author: Chen, Yanjun
• Other Authors: Yap Kim Hui
• Format: Final Year Project
• Language: English
• Published: 2016
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/67706
• Institution: Nanyang Technological University

Visual Product Search System in Mobile Commerce consists of the mobile-side application for customer users to scan as well as to search and the web-side application for service subscribers to create and manage the product database. In the mobile-side, the application performs search through image recognition and matching. After clients query a search, the query image will be extracted to some meaningful descriptors by the SIFT algorithm which will be used for matching after then. For image matching, the BoW (Bag of Words) methodology is used. This method is used to generate a large codebook, which is called vocabulary tree by k-means clustering. All the descriptors coming in will go through this vocabulary tree and the highest scored images will be the candidates returned. Geometric verification is used after this featuring matching to find the mostly matched result from the candidates generated above. Image and corresponding information will be retrieved from database and server to be returned back to clients according to the mostly matched result. In the web-side, the application works as a platform of the database system management for the service subscribers which means those business runners. In this project, the enhanced web application allows users from different business fields to execute the management. By using forms in the web site, the clients can define the unique attributes of their products to create their own product tables. By using the uploading and updating functionalities, the clients can upload and update both single and multiple incremental images and related information into database and server. For every business runner, the database can work especially for them to store their data. By integration, the whole visual product search system in mobile commerce is a multi-domain, robust and efficient system for both the customers who can get more interested information by using the mobile application and the subscribers who can manage the database to run a better business by using the web application. The realistic tests proved the efficiency and 100% accuracy of this system. The results verified the capabilities of all functionalities of the web application and the whole system.