Authors - Jobin Francis, Akhil Johnson, Baburaj Madathil, Sudhish N George Abstract - Traditional clustering methods, which vectorize the images for clustering, frequently fail to consider the intrinsic structure of imaging data. Hence, a tensor-based clustering framework is proposed, which leverages the self-expressiveness property of submodules to preserve the spatial characteristics of images. To capture better low rankness and self-expressiveness property, an l1/2-induced Tensor Nuclear Norm (TNN) is proposed. In the submodule structural constraint, l1/2 regularisation is employed because of its inherent noise robustness and the ability to provide a more sparser solution. An optimization problem is formulated using the capabilities of l1/2-induced TNN and l1/2 regularization. Three popular datasets are used to evaluate the performance of proposed method. The results show that proposed method shows improved clustering results than the state-of-the-art methods compared.
Open Zoom