On improving robustness of video fingerprints based on projections of features
Regunathan Radhakrishnan, Claus Bauer · 2009
In this paper, we study two methods to improve the robustness property of projection based hashing methods. For this class of hashing methods, a feature matrix is projected onto a set of projection matrices. Then, the projected values are compared to a threshold to derive the hash bits. In our previous work [4], we showed that the collision characteristics of these methods can be optimized by a careful selection of the projection matrices. The projection matrices were obtained using a Singular Value Decomposition (SVD) on a set of features from a training data set that minimized the cross-correlation between projected values. However, these projection matrices did not consider the effect of content modifications on the projected values