Mobile visual recognition

• Main Author: Aung, Thandar
• Other Authors: Yap Kim Hui
• Format: Final Year Project
• Language: English
• Published: 2017
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/72185
• Institution: Nanyang Technological University

Mobile visual search (MVS) is emerged due to the increasing use of smart phones nowadays. It allows the users to search any information via smart phones using images instead of solely texts. Thus, the main advantage of MVS is the ease of user to explore many fields like food browsing or shopping.The project scopes focused by past FYP students are mainly based on objects that have definite shapes to test out the robustness of the algorithm of MVS. However, there are varieties of objects which have unique shapes at different times. Therefore, the main objective of this project is to test out with 3D deformable shapes to evaluate the performance of the MVS. Then, the image database of 3D deformable snack packages is constructed using images taken by Iphone 7 plus for both the reference images and the test images. Comparison and evaluation on different methods will be done in terms of the overall accuracy of the system. Limitations exist such as unique shapes of deformable snack packages and unwanted features such as background noise and occlusions by the users. Those will cause as “Noise” in image recognition and will cause the overall accuracy percentage to drop significantly. Hence, Geometric Verification (GV) method has been introduced and it will help to improve the overall matching accuracy by 5-10%. However, the required time to process with GV is much longer in the image recognition process. We will conduct different sets of experiments to evaluate the final results with and without GV. Lastly, we will discuss further on the improvement of the image matching accuracy and database optimization.