Overview
Image feature descriptors play crucial roles in image matching or recognition tasks. For mobile devices to perform these tasks, the major challenge was to exhibit robust matching performance while at the same time maintain the reasonable storage requirement. To achieve this, I proposed Dynamic Local Intensity Order Relations (DLIOR) descriptor by considering pairwise intensity relation in a pixel group. As a result, my method reduced storage space successfully while reaching comparable feature matching performance.
Role
I proposed an optimized algorithm of local image descriptor and developed it with C++ and OpenCV.
Advisor
Wen-Hung Liao
Tools
C++ & OpenCV
Publication
In Proceedings of ICIAR'19, Aug, 2019 | [PDF] | [Poster]
Image Matching Process
Result
DLIOR descriptor outperforms the state-of-the-art method LIOP descriptor by 1.5% in image matching task and reduced storage space from N! to CN2 .
