A novel DSSP (double signal subspace projection) algorithm for better VD (Virtual Dimensionality) estimation

Sections 1 and 2 of the full paper explain and evaluate our DSSP algorithm as mentioned in the title; we believe it is novel and its VD estimation is better. The core of sections 1 and 2 consists of: (1). the pixel representation and image representation of a hyperspectral image are utilized to generate respectively two sets of sub-spaces according to the principal component analysis (PCA); (2). when the dimensionality of these two sets of subspaces exceeds VD of the image, both subspace projections show the same reconstruction performance; therefore, VD can be estimated by judging the difference between reconstruction performances of these two subspace projections in DSSP instead of distinguishing signal subspace from noisy subspace; (3). the results of both synthetic and real hyperspectral experiments, given in Figs. 1 through 9, demonstrate preliminarily that the performance of traditional signal subspace projection based VD estimation has been improved and that the performance of the proposed DSSP based VD estimation algorithm outperforms those of the noise whitened HFC (NWHFC) and signal subspace estimation (SSE) based VD estimation algorithms.