Towards more discriminative features for texture recognition
作者：
Highlights：
• We propose a novel algorithm for extracting more discriminative 2D LBP features.
• The method optimizes projections of 2D LBP distribution on to the marginal histograms.
• Several statistical constraints are considered for optimization.
• Mutual Information constraint has shown the best performance among others.
• The method shows better performance up to 2.4% compared to conventional approach.
摘要
•We propose a novel algorithm for extracting more discriminative 2D LBP features.•The method optimizes projections of 2D LBP distribution on to the marginal histograms.•Several statistical constraints are considered for optimization.•Mutual Information constraint has shown the best performance among others.•The method shows better performance up to 2.4% compared to conventional approach.
论文关键词：Local binary patterns,Texture recognition,Feature representations,Feature optimization,Deep texture features,Mutual information
论文评审过程：Received 11 February 2019, Revised 21 May 2020, Accepted 23 May 2020, Available online 30 May 2020, Version of Record 9 July 2020.
论文官网地址：https://doi.org/10.1016/j.patcog.2020.107473
