Learning Image and Video Representations Based on Sparsity Priors
Xian Wei · 2017
Sparse representations of data have been observed to contain rich distributed information of the data with respect to specific learning tasks, such as image classification, regression, etc. By taking advantage of such a benefit, the focus of this dissertation is on developing a two-layer representation learning framework that allows disentangling the underlying explanatory factors hidden in sparse representations of image and video data. Such a learning framework has been successfully applied on modeling dynamic textures and finding low dimensional image representations.