Dynamic texture classification method based on Spectral Time-Vertex Wavelet Transform

 Dynamic textures are sequences of images that change over time. Dynamic texture classification plays an important role in medical testing, industrial production and forest fire control. Dynamic textures exhibit “appearance” and “motion” properties in space and time. Combining these two properties, a dynamic texture classification method based on spectral time-vertex wavelet transform and Marginal distribution covariance model was proposed in this paper. The time-vertex graph signal processing framework was used to represent the dynamic texture as the time-vertex graph signal. Since Meyer wavelet can represent dynamic textures in multiple directions and at multiple scales, so the multi-scale decomposition of dynamic texture was performed by spectral time-vertex Meyer wavelet transform. Then, Marginal distribution covariance model was applied to each sub-band, and the characteristic covariance matrix of intra-band correlation was obtained as dynamic texture feature for classification. Due to the representation of time-vertex graph can effectively describe the spatial relations among dynamic texture pixels and their changes along time, meanwhile, spectral wavelet transform inherits the advantages of graph representation and wavelet transform, so we used spectra time-vertex wavelet decomposition and marginal distribution covariance model to obtain dynamic texture feature effectively. The experiment results on standard dynamic texture data sets show that the proposed method has good classification performance.