FIGURE 1.

The overall architecture of the proposed generative adversarial network constrained multiple loss autoencoder (GANCMLAE). Encoder1, Decoder, Encoder2, Discriminator: the based structure; L1, L2, L3: the constituent elements of the combined loss; x: the input of the model; z: the generated latent vector of x, negative training data for the discriminator; x′: the reconstructed image; z′: the generated latent vector of x′. The vector sampled from the normal distribution: the positive training data for the discriminator