Mobile visual product search

• Main Author: Heng, Yann Wu
• Other Authors: Yap Kim Hui
• Format: Final Year Project
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/77401
• Institution: Nanyang Technological University

In recent years, smartphone development has been exploding. Nowadays we have smartphones with multiple in-built camera modules, smartphones with 5G connectivity, smartphones with storage as big as 1 terabyte and many other very cool features. With all these features given to users, unlike previously where phones are only for communication, now we can do more things and achieve many great things with only using our phone. One of the examples is to take images using smartphone and all the relevant information the user required are shown to the users within seconds. Mobile visual search is the technology that we are going to discussed in this project. This project is to let users take images of their books cover and retrieve information regarding the books. To study the method and algorithm, we used Matlab code to simulate the image taking and analysing process and whereby database of training and testing were constructed to simulate the background environment. Several experiments were conducted based on books cover images under normal condition, with occlusions and with background noise. Other than that, Geometric Verification was used to compare the accuracy and time taken to process all the images in the database. From the results obtained, we know that images with occlusion have poorer image recognition accuracy as compared to images under normal condition and images with background noise. Once incorporated with GV, the image recognition accuracy was increased, but this is done with the sacrifices of time taken to process all the images. Finally, future works to incorporate Speeded-Up Robust Features (SURF) instead of Scale-Invariant Feature Transform (SIFT) was proposed to improve the time taken to process all the images as the databases is constantly growing.