Algorithm 2 Pseudo-code of rotation angle refinement matching
1:	Inputs: Test image I, size of template w × h, coordinate candidate ($x_{Cand2}$, $y_{Cand2}$), angle
	candidates ${\theta }_{Cand2},$ scale candidates $s_{Cand2}$, angular resolution α, number of candidates $N_{Cand2}$
2:	Outputs: Correlation coefficient ${\eta }_{NCC}$
3:	X index: idxX$\leftarrow$blockDim.x ∗ blockIdx.x + threadIdx.x   //Angular resolution
4:	Y index: idxY$\leftarrow$blockDim.y ∗ blockIdx.y + threadIdx.y   //Number of candidates
5:	if (idxX < $\mathsf{\alpha }\ast 10$) and (idxY < $N_{Cand2}$) then
6:	Angle $\theta$: ${\theta }_{Refined}$ $\leftarrow \left(idxX / 10.0\right)+{\theta }_{Cand2}$[idxY]   //Refined angle
7:	Coordinate X: $x\leftarrow x_{Cand2}$[idxY]
8:	Coordinate Y: $y\leftarrow$ $y_{Cand2}$[idxY]
9:	Scale s: s $\leftarrow s_{Cand2}$[idxY]
10:	for j in h do
11:	for i in w do
12:	Collect intensity pixel values inside a search window using a bilinear interpolation approach with a center point at (x, y), an orientation: ${\theta }_{Refined}$, and a scale: s
13:	end
14:	end
15:	Calculate the NCC score ${\eta }_{NCC}$ between the template and the search windowend
16:	end
17:	Return${\eta }_{NCC}$