Algorithm 4. RANSAC method to best subset image selection.

input: $X=[x_1,x_2,\dots ,x_n]$

Initialization: r1 as restored image by averaging on m restored images, r2 as $(m+1{)}^{th}$ restored image

for k = 1:C

select mi images randomly

for $l=1:n-m$

calculate r1: apply Algorithm 5 on mi images

calculate r2: apply Algorithm 5 on mi images in addition to each remaining image $\{g_j=1,\dots ,n-m\}$

calculate C1j: calculate similarity of r1 and r2 images using cross correlation

calculate C2j: calculate noise distribution distance of m and m + 1 image sets using KL divergence

calculate C3j: calculate transformation distribution distance of m and m + 1 image sets using KL divergence

add gi image to support set of mi images if weighted average of $C{1}_j,C{2}_j,C{3}_j$ is less than predefined threshold

end for

end for

select $\{m^i=1,\dots ,C\}$ that have largest support set as m* and related support set as s*

$return\{m\ast +s\ast \}$