Dynamic texture representation based on oriented magnitudes of Gaussian gradients
Thanh Tuan Nguyen1,2, Thanh Phuong Nguyen1 and Frederic Bouchara1
1Aix Marseille University, University of Toulon, CNRS, LIS, UMR 7020, 13397 Marseille, France
2HCMC University of Technology and Education, Faculty of IT, HCM City, Vietnam
Abstract
Efficiently capturing shape and turbulent motions of dynamic texture (DTs) for video description is a challenge in real applications due to the negative influences of the well-known problems: environmental elements, illumination, scale, and noise.
In this paper, we propose an efficient and simple framework for DT representation based on oriented features of high-order Gaussian gradients. Firstly, 2D/3D Gaussian-based filtering kernels in high-order partial derivatives are taken into account video analysis as a preprocessing to obtain corresponding gradient-filtered images/volumes. After that, oriented features, which are robust against above issues, are extracted by decomposing the Gaussian derivative magnitudes into oriented components.
Finally, a shallow local encoding is utilized for structuring spatio-temporal features from these oriented magnitudes. This allows to construct discriminative descriptors with promising performances compared to those based on the non-oriented ones.
Experimental results for DT classification task on benchmark datasets have verified the interest of our proposal.
Our code
Here is MATLAB code of our modified soft assignment to decompose
high-order 2D/3D Gaussian gradients subject to a pre-defined orientation range
If you use this code in your work, please cite the paper [1]
References
[1] Thanh Tuan Nguyen, Thanh Phuong Nguyen, Frederic Bouchara, Dynamic texture representation based on oriented magnitudes of Gaussian gradients, Journal of Visual Communication and Image Representation, 2020 (submission)