Please use this identifier to cite or link to this item:
|Title:||Mobile visual search and services||Authors:||Zhao, Heng||Keywords:||DRNTU::Engineering||Issue Date:||2016||Abstract:||Mobile Visual Search (MVS) is a newly emerged concept, which uses smart phone as a platform to provide various services through image search. MVS has some natural advantages, its unique search method and easiness of use proved to be useful and promising for applications in many fields, such as shopping, browsing, etc. However, most of the MVS applications need to capture and store the image before recognition, which limits the visual search to only one image per cycle. As a result, it slows the pace of searching thus may lower user experience. Thanks to Augmented Reality (AR), this gap could be solved by integrating AR technique into existing applications. By this way, user could search different objects and check the results on the camera screen without leaving the page. Furthermore, AR technique uses image scanning instead of capturing, which leads to a faster and more interactive result display. Under this project, the goal is to develop such a MVS application together with AR technique within the scope of book recognition, in order to help user obtain book information and provide a convenient way of browsing and purchase. The application allows user to switch between ‘auto-scanning’ mode and ‘manually add’ mode to add interested books, and then the system will generate a list based on price or rating to help user sort them. User can access books’ information and recommendations via the following interface. At the end of the development phase, a series of systematic tests were conducted to verify the performance of the application. On three different Android platforms, the application achieved an accuracy up to 97.7%, and a response time of 1.2s on average. The effects of some environmental variables were also tested. The database involved contained 130 books in total. It was also proved that the accuracy could be enhanced further by changing the threshold value for recognition process, the new accuracy data reached 98.7% with a response time averaged to 1.25s.||URI:||http://hdl.handle.net/10356/67561||Rights:||Nanyang Technological University||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||EEE Student Reports (FYP/IA/PA/PI)|
Page view(s) 1096
Updated on Nov 25, 2020
Updated on Nov 25, 2020
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.