Algorithm 1 The proposed dictionary learning algorithm

Inputs: Two group of images $\{{\tilde{X}}_{d,l}\}$ ($l=1,2$)

(1) Extract patches of $\{{\tilde{X}}_{d,1}\}$ and $\{{\tilde{X}}_{d,2}\}$ from upper left to lower right.

(2) Patch classification based on SNE and MSF in Equations (6) and (8), respectively.

(3) Construct two training sets $T_e$ and $W_c$ by Equations (13) and (15), respectively.

(4) Obtain the two sub-dictionaries $D_e$ and $D_c$ by solve Equations (17) and (19), respectively.

(5) Generate the final dictionary $D$ by Equation (20)

Output: The overcomplete dictionary $D$