13 March 2013 Real-time billboard trademark detection and recognition in sports video
Author Affiliations +
Abstract
Nowadays, different applications like automatic video indexing, keyword based video search and TV commercials can be developed by detecting and recognizing the billboard trademark. We propose a hierarchical solution for real-time billboard trademark recognition in various sports video, billboard frames are detected in the first level, fuzzy decision tree with easily-computing features are employed to accelerate the process, while in the second level, color and regional SIFT features are combined for the first time to describe the appearance of trademarks, and the shared nearest neighbor (SNN) clustering with x2 distance is utilized instead of traditional K-means clustering to construct the SIFT vocabulary, at last, Latent Semantic Analysis (LSA) based SIFT vocabulary matching is performed on the template trademark and the candidate regions in billboard frame. The preliminary experiments demonstrate the effectiveness of the hierarchical solution, and real time constraints are also met by our solution.
© (2013) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Jiang Bu, Jiang Bu, Song-Yan Lao, Song-Yan Lao, Liang Bai, Liang Bai, } "Real-time billboard trademark detection and recognition in sports video", Proc. SPIE 8783, Fifth International Conference on Machine Vision (ICMV 2012): Computer Vision, Image Analysis and Processing, 87831M (13 March 2013); doi: 10.1117/12.2021184; https://doi.org/10.1117/12.2021184
PROCEEDINGS
8 PAGES


SHARE
Back to Top