Hybrid Resolution Spectral Imaging by Class-based Regression Method

摘要

Hybrid resolution spectral imaging produces spectral imagesrnfrom high-resolution RGB images and corresponding lowresolutionrnspectral data. Various methods have been proposed,rnwhereas the low-resolution spectral data are regarded as thernsample data of target scenes. However, this approach is notrnappropriate when each spectrum in the low-resolution data may berna mixture of spectra with different spectral features, and thernoriginal spectral feature is lost by averaging them. To solve thisrnproblem, class-based regression method for mixed low-resolutionrnspectral data was proposed. In this method, the spectral estimationrnmatrix for every class is derived using a regression approach,rnwhere the clustering results of the high-resolution RGB image arernused to incorporate spectral unmixing. However, the method wasrntested only for small regions of images. In this paper, spectralrnimages are estimated by the class-based regression method forrnthree test spectral images, and the accuracy is compared with twornconventional methods for hybrid resolution spectral imaging.rnExperiments confirm that the spectra are accurately reconstructedrnonly by class-based regression method when they are observed asrnmixed spectra in the low-resolution data.