Skip to main content
EURASIP Journal on Image and Video Processing

Table 1 Network architecture of proposed model

From: The secure steganography for hiding images via GAN

Encoder

Decoder

Discriminator (Xu’Net)

Input:6×256×256

Input:3×256×256

Input:3×256×256

Conv(6, 64, 7)

  

Conv(64,128,3)

  

Conv(128,256,3)

  

Res_block1(256,256,3)

Conv(3, 64,3)

kv_filter(3,1,5)

Res_block2(256,256,3)

Conv(64, 128,3)

Conv(1, 8, 5)

Res_block3(256,256,3)

Conv(128,256,3)

Conv(8, 16, 5)

Res_block4(256,256,3)

Conv(256,128,3)

Conv(16, 32, 3)

Res_block5(256,256,3)

Conv(128, 64, 3)

Conv(32, 64,3)

Res_block6(256,256,3)

Conv(64, 3,3)

Conv(64,128,3)

Res_block7(256,256,3)

Sigmoid(3, 3)

Sigmoid(128,1)

Res_block8(256,256,3)

  

Res_block9(256,256,3)

  

DeConv(256,128, 3)

  

DeConv(128, 64, 3)

  

Conv(64, 3, 7)

  

Tanh(3,3)

  

Output:3×256×256

Output:3×256×256

Output:1

  1. Each resnet block includes two 3×3 convolutional layers with batch normalization operation and ReLU activation function, which outputs the result of adding the input feature map and the feature map after two convolutional layers
Back to article page