Hybrid resolution spectral imaging produces spectral images from high-resolution RGB images and corresponding low-resolution spectral data. Various methods have been proposed, whereas the low-resolution spectral data are regarded as the sample data of target scenes. However, this approach is not appropriate when each spectrum in the low-resolution data may be a mixture of spectra with different spectral features, and the original spectral feature is lost by averaging them. To solve this problem, class-based regression method for mixed low-resolution spectral data was proposed. In this method, the spectral estimation matrix for every class is derived using a regression approach, where the clustering results of the high-resolution RGB image are used to incorporate spectral unmixing. However, the method was tested only for small regions of images. In this paper, spectral images are estimated by the class-based regression method for three test spectral images, and the accuracy is compared with two conventional methods for hybrid resolution spectral imaging. Experiments confirm that the spectra are accurately reconstructed only by class-based regression method when they are observed as mixed spectra in the low-resolution data.
Yuri Murakami, Masahiro Yamaguchi, Nagaaki Ohyama, "Hybrid Resolution Spectral Imaging by Class-based Regression Method" in Proc. IS&T 19th Color and Imaging Conf., 2011, pp 310 - 315, https://doi.org/10.2352/CIC.2011.19.1.art00059