|Title:||Human-themed image background defocusing on mobile devices with deep learning approaches|
|Advisors:||Zhang, Lei (COMP)|
|Subject:||Hong Kong Polytechnic University -- Dissertations|
Image processing -- Digital techniques
|Department:||Department of Computing|
|Pages:||vi, 47 pages : color illustrations|
|Abstract:||The system, for defocusing the background of human-themed images, is one that emphasizes people as foreground and blurring other objects as background. As mobile devices become more and more pervasive to daily life of humankinds, the background defocusing on human-themed images turn to be essential in many mobile platforms (smartphone, UAV, etc.). Within this thesis, the author designed and implemented a prototype system for defocusing the background of human-themed images on mobile devices. The aim of the system is providing a solution to generate "Bokeh" effects, which are commonly seen on SLR (Single Lens Reflex) cameras, from human-themed pictures on mobile devices (e.g. smartphones). The system discussed in this thesis mainly includes three parts for the three subtasks of image defocusing: portrait segmentation, depth estimation, image fusion and defocusing. For the first parts, the author applied a decoder-encoder structure based on DeepLab V3+ with a light-weight encoding backbone MobileNet V2. For the second part, the author applied a depth map generating method based on DenseNet. For the last part of the system, the author applied Gaussian blurring in different magnitude with the information of relative depth. As the experiments conducted on PC, the results showed that this prototype could provide state-of-art performance. The author also transplanted the prototype system to Android devices (specifically, an Android smartphone) and the experiment result showed that it cost less than one second per image on mobile devices, which means that the prototype is appropriate for being deployed on mobile devices.|
|Rights:||All rights reserved|
Files in This Item:
|991022268444103411.pdf||For All Users (off-campus access for PolyU Staff & Students only)||1.99 MB||Adobe PDF||View/Open|
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item: