Algorithm 1. Two layer perceptual image fusion using deep learning.

Procedure: Fusion Schema using NSCT

Input $\leftarrow$ X, Y are two CXR input image(s) and VGGNet is the pretrained VGG-16 network.

Output $\leftarrow$ Z is the fuzed CXR image

Begin

Xa, Xb $\leftarrow$ decomposeByNSCT (X)

Ya, Yb $\leftarrow$ decomposeByNSCT (Y)

Za $\leftarrow$ DeepLearningFusionRule (Xa, Ya, VGGNet)

Zb $\leftarrow$ DeepLearningFusionRule (Xb, Yb, VGGNet)

Z $\leftarrow$ recomposeByNSCT(Za,Zb)

saveImageFile(Z)

[ ] $\leftarrow$ evalautionMetrics(X, Y, Z)

Print([ ])

End

Procedure Deep Learning Fusion Rule

Input $\leftarrow$ A, B are two subband of CXR input image(s) and VGGNet is the pretrained VGG-16 network.

Output $\leftarrow$ C is the fuzed subband of CXR image

Begin

$A_{feature}$ $\leftarrow$ extractFeatures(VGGNet, A)

$A_{vec}\leftarrow A_{feature}$

$A_{TC}\leftarrow A_{vec}/norm\left(A_{feature}\right)$

For each $B_{}i$ in B subband CXR images

$B{i}_{feature}$ $\leftarrow$ extractFeatures(VGGNet, $B_{}i$)

$B{i}_{vec}\leftarrow B{i}_{feature}$

$B{i}_{TC}\leftarrow B_{vec}/norm\left(B_{feature}\right)$

End

$S,V_{img},denom\leftarrow \mathbf{z}\mathbf{e}\mathbf{r}\mathbf{o}\mathbf{s}\left(B\right)$

For each $B_{}i$ in B subband CXR images

$S\leftarrow \mathbf{S}+B{i}_{TC}$

$V_{img}\leftarrow V_{img}+B{i}_{vec}$

$denom\leftarrow denom+B{i}_{vec}$

End

For each $B_{}i$ in B subband CXR images

$W\leftarrow B{i}_{TC}/denom$

End

$C\leftarrow \mathbf{z}\mathbf{e}\mathbf{r}\mathbf{o}\mathbf{s}\left(B\right)$

For each $W_{}i$ in W

$C\leftarrow C+B_{}i\ast W_{}i$

End

End