Table 2.

Pseudo-codes for the proposed algorithm.

Initialization

Setting αk,i,1, αmax, c1, c2, lmax, D, kmax;

Initializing f⃑0, k = 0, τ = 1;

Obtaining an RI u⃑;

Reconstruction procedure (for any channel)

while k < kmax

k + +;

Updating f⃑k−1 to ${\overline{f}}_{OS-SART}^k$ using the OS-SART method;

Initializing the patch index i;

Repeat (until patch window moves over entire image)

Subtracting the DC component to get ${\overline{f}}_{OS-SART}^{k.i}$ and ũi;

Determining the update direction p⃑k,iwith ${\overline{f}}_{OS-SART}^{k.i}$ and ũi;

Determining the step length αk,i,* using ${\overline{f}}_{OS-SART}^{k.i}$, ũi and p⃑k,i according to conditions (13) and (15);

Updating ith patch using equation (10);

Normalizing the patch and adding back the DC component to get ${\overline{f}}_{OS-SART}^{k.i}$;

i + +;

end (Repeat)

Averaging different values for the same pixel to get the image estimate f⃑k+1;

Accelerating the result by using FISTA;

Calculate dk;

If k mod T = 0, ${\overline{d}}^{\tau }=\frac{1}{T}{\sum }_{j=k-T+1}^k{d}^k$;

If the difference of d̄τ and d̄τ−1 is small enough, break;

τ + +;

end(if)

end (while)

Output:f⃑k+1