Spectral image clustering discovers groups and identifies distributions from spectral signatures without requiring a previous training stage. The sparse subspace clustering-based methods group spectral signatures in different subspaces, finding the sparsest representation for each pixel, guaranteeing that they belong to the same class. Although these methods have shown good accuracy, as the number of pixels increases, computational complexity becomes intractable. This paper proposes to reduce the number of pixels to be classified in the spectral image by half through a sub-sampling procedure that eliminate every two contiguous pixels, preserving the spatial structure of the image. Then, the clustering result is obtained using its spatial information. The performance of the proposed spectral image clustering framework is evaluated in three datasets showing that similar accuracy is obtained up to 7.9 times faster compared to the classification of the data sets without subsampling.