Convolutional neural network on three orthogonal planes for dynamic texture classification

作者：

Highlights：

• A new CNN framework is introduced to analyze DTs on three orthogonal planes.

• Multiple texture specific CNNs are developed.

• An analysis of the contribution of each plane is conducted as well as the domain transferability.

• Experiments on various DT classification datasets show the superiority of our approach over existing ones.

摘要

•A new CNN framework is introduced to analyze DTs on three orthogonal planes.•Multiple texture specific CNNs are developed.•An analysis of the contribution of each plane is conducted as well as the domain transferability.•Experiments on various DT classification datasets show the superiority of our approach over existing ones.

论文关键词：Dynamic texture,Image recognition,Convolutional neural network,Filter banks,Spatiotemporal analysis

论文评审过程：Received 4 April 2017, Revised 30 September 2017, Accepted 23 October 2017, Available online 24 October 2017, Version of Record 21 December 2017.

论文官网地址：https://doi.org/10.1016/j.patcog.2017.10.030