A novel filtering kernel based on difference of derivative Gaussians with applications to dynamic texture representation
作者：
Highlights：
• A novel DoDG kernel is introduced to deal with negative impacts on DT description.
• An investigation has been made to verify eminent efficiency of DoDG compared to DoG.
• DoDG is considered in multi-orders to exploit more high-order DoDG-filtered features.
• A shallow framework is introduced to take DoDG into account local analysis of videos.
• Our DoDG-based descriptors obtain very good performance compared to state of the art.
摘要
•A novel DoDG kernel is introduced to deal with negative impacts on DT description.•An investigation has been made to verify eminent efficiency of DoDG compared to DoG.•DoDG is considered in multi-orders to exploit more high-order DoDG-filtered features.•A shallow framework is introduced to take DoDG into account local analysis of videos.•Our DoDG-based descriptors obtain very good performance compared to state of the art.
论文关键词：Dynamic texture,Feature extraction,Gaussian-based filterings,LBP,CLBP,Video representation
论文评审过程：Received 14 December 2020, Revised 3 June 2021, Accepted 13 July 2021, Available online 19 July 2021, Version of Record 6 August 2021.
论文官网地址：https://doi.org/10.1016/j.image.2021.116394
