Quality Evaluation of Hyperspectral Image Denoising Algorithm Based on Classification Application

摘要

Signal-to-noise ratio (SNR) is a widely accepted image quality evaluation criterion for its easy calculation and explicit mathematical significance. However, for quality evaluation of hyperspectral image denoising algorithms, SNR ignores the structure of the spatial information of hyperspectral image, and does not represent the loss of the spectral information. Therefore, we use classification to assess the hyperspectral image denoising algorithms. Three hyperspectral image denoising methods are implemented respectively, which are: 1) hyperspectral image denoising base on principal component analysis (PCA); 2) multiple linear regression (MLR) model for hyperspectral image denoising in the wavelet domain; 3) MRL model for hyperspectral image denoising in the curvelet domain. The classification results show that the higher SNR does not always contribute to higher overall accuracy of classification. Therefore, the hyperspectral image quality assessment based on the post application is necessary.