This paper presents a novel feature extraction model that incorporates local histogram in spatial space and pixel spectrum in spectral space, with the goal of hyperspectral image classification. We named this joint spectrum as 3D spectrum. Moreover, as a pre-processing step, an iterative procedure, which exploits spectral information in such a way that it considers corrupted bands existing in the data cube, is applied to original hyperspectral image. Further, Affine transform is applied to the bands chosen by the aforementioned procedure. The final feature is extracted by affine transform and 3D spectrum model, and as an input of widely used classifier of Support Vector Machine. As a post-processing step, multiple iterative results are fused in the level of probability. Our experimental results indicate that the proposed methodology leads to state-of-the-art classification results when combined with probabilistic classifiers for several widely used hyperspectral data sets, even when very only limited training samples are available. © 2018, Springer Science+Business Media, LLC, part of Springer Nature.