Learning graph representation with Randomized Neural Network for dynamic texture classification

Lucas Correia Ribas, Jarbas Joaci de Mesquita Sá Junior, Antoine Manzanera, Odemir M. Bruno

Abstract: Highlights•Dynamic texture recognition method is proposed.•It combines a graph-based description and Random Neural Network (RNN) model.•Dynamic texture video is modeled as a space ×<math><mo is="true">×</mo></math> time graph.•Local topological features are used to train the RNN.•Results on Dyntex++ and UCLA datasets present the method performance.