Algorithm 1. Generating sparse matches.

Input: I1, I2, a threshold ς

Output: a set $\mathcal{M}$ of sparse patch matches

1:   Divide the image I1 into a set of small regions $\{{\mathrm{R}}_1,{\mathrm{R}}_2,\dots ,{\mathrm{R}}_M\}$

2:  for each ${\mathrm{R}}_i\in I_1$ do

3:   for any region ${\mathrm{P}}_j\in I_2$ do

4:     set $D({\mathrm{R}}_i,{\mathrm{P}}_j)=BBS({\mathrm{R}}_i,{\mathrm{P}}_j)$

5:     set $\mathcal{P}\mathcal{M}({\mathrm{R}}_i,{\mathrm{P}}_j)=\{(r_i,p_j)|r_i\in {\mathrm{R}}_i,p_j\in {\mathrm{P}}_j,S(r_i,p_j)=1\}$

6:   end for

7:  end for

8:  Determine the optimal matching set ${\mathcal{M}}_c=\{({\mathrm{R}}_i,{\mathrm{R}}_i^{\mathrm{\prime }})|{\mathrm{R}}_i^{\mathrm{\prime }}=\underset{{\mathrm{P}}_j\in I_2}{\arg max}D({\mathrm{R}}_i,{\mathrm{P}}_j)\}$

9:  for each $({\mathrm{R}}_i,{\mathrm{R}}_i^{\mathrm{\prime }})\in {\mathcal{M}}_c$ do

10:  if $BBS({\mathrm{R}}_i,{\mathrm{R}}_i^{\mathrm{\prime }})>\varsigma$ then

11:    $\mathcal{M}=\mathcal{M}\bigcup \mathcal{P}\mathcal{M}({\mathrm{R}}_i,{\mathrm{R}}_i^{\mathrm{\prime }})$

12:  end if

13:   end for