Abstract
In this paper, we propose dynamic texture recognition from video snippets by constructing temporal intensity co-occurrence histograms for feature representation and learning. The pair-wise intensity co-occurrence frequencies are summarized from every pixel position between every pair of sequential frames in the video separated by a certain time lapse or offset distance. A 256 × 256 grayscale intensity co-occurrence matrix is thus constructed for the given offset distance. Twenty offset distances from d = 1, 2, …, 20 are used for the computation that yields twenty 256 × 256 temporal co-occurrence matrices from a single video. The twenty 2D histograms so formed are individually converted to complete probability distributions whose elements sum up to one, and then each histogram is converted to a 1D feature vector. The twenty 1D feature vectors represent local patterns that are concatenated to form a unique ID pattern that is matched using the ensemble of bagged decision trees classifier. Alternatively, a 20 × 20 grid of distance classifiers is substituted to find matches between the local patterns followed by the summation of distances from all the grids. The recognition rate achieved in our experiments is found superior to the state-of-the-art, when tested on the raw, unprocessed, and unsegmented videos of the benchmark Dyntex++ dataset.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.