This paper proposes a new hyperspectral image subspace clustering framework which adds a total variation denoising constraint in order to improve the similarity between data points from the same subspace.