Efficient feature description and classification of dynamic texture (DT) is an important problem in computer vision and pattern recognition. Recently, the local binary pattern (LBP) based dynamic texture descriptor has been proposed to classify DTs by extending the LBP operator used in static texture analysis to the temporal domain. However, the extended LBP operator cannot characterize the intrinsic motion of dynamic texture well. In this paper, we propose a novel video set based collaborative representation dynamic texture classification method. First, we divide the dynamic texture sequence into subsequences along the temporal axis to form the video set. For each DT, we extract the video set based LBP histogram to describe it. We then propose a regularized collaborative representation model to code the LBP histograms of the query video sets over the LBP histograms of the training video sets. Finally, with the coding coefficients, the distance between the query video set and the training video sets can be calculated for classification. Experimental results on the benchmark dynamic texture datasets demonstrate that the proposed method can yield good performance in terms of both classification accuracy and efficiency.
- Collaborative representation
- Dynamic texture classification
- Local binary pattern
- Texture feature extraction
ASJC Scopus subject areas
- Signal Processing
- Computer Vision and Pattern Recognition