Video Google: A Text Retrieval Approach to Object Matching in Videos
ICCV '03 Proceedings of the Ninth IEEE International Conference on Computer Vision - Volume 2
Scale & Affine Invariant Interest Point Detectors
International Journal of Computer Vision
Distinctive Image Features from Scale-Invariant Keypoints
International Journal of Computer Vision
k-means++: the advantages of careful seeding
SODA '07 Proceedings of the eighteenth annual ACM-SIAM symposium on Discrete algorithms
Image retrieval: Ideas, influences, and trends of the new age
ACM Computing Surveys (CSUR)
| Hi-index | 0.00 |
This contribution presents a system for detection of similar images of advertisements in moderate size datasets. These datasets are daily updated and mainly consists of advertisements from tv, newspapers, journals, etc. The task is to identify clusters of duplicate advertisements in given dataset. Images differ by translation, scale and the amount of compression. The presented approach is based on recently popular bag-of-features approach which has been successfully used in context of image retrieval and other related areas. Each image is represented as weighted histogram of local features. Similarities are computed based on the extracted features are projected onto separating hyperplane and clustered using agglomerative hierarchical clustering. Experiments show that this simple and efficient scheme yields good results and finds corresponding images even for advertisements which are substantially dissimilar in spatial arrangement and color composition with reasonable false positive rate.