Sequential dimensionality reduction for extracting localized features
作者：
Highlights：
• We propose a new nonnegative matrix underapproximation model for images.
• We design an algorithm that runs in linear time in the dimensions of the input matrix.
• This allows us to extract sequentially localized and spatially coherent features.
• We illustrate the effectiveness of our approach on a synthetic data set, facial and hyperspectral images.
• We show that it competes favorably with comparable state-of-the-art techniques.
摘要
Highlights•We propose a new nonnegative matrix underapproximation model for images.•We design an algorithm that runs in linear time in the dimensions of the input matrix.•This allows us to extract sequentially localized and spatially coherent features.•We illustrate the effectiveness of our approach on a synthetic data set, facial and hyperspectral images.•We show that it competes favorably with comparable state-of-the-art techniques.
论文关键词：Nonnegative matrix factorization,Underapproximation,Sparsity,Hyperspectral imaging,Dimensionality reduction,Spatial information
论文评审过程：Received 1 June 2015, Revised 4 July 2016, Accepted 10 September 2016, Available online 19 September 2016, Version of Record 28 September 2016.
论文官网地址：https://doi.org/10.1016/j.patcog.2016.09.006
