Adaptive Deep Supervised Autoencoder Based Image Reconstruction for Face Recognition

Figure 3

Architecture of SSAE and ADSNT. (a) Supervised autoencoder (SAE) which is comprised of clean/“corrupted” datum, one hidden layer, and one reconstruction layer by using the “corrupted datum”; (b) stacked supervised autoencoder (SSAE); (c) architecture of the Adaptive Deep Supervised Network Template (ADSNT).

(a)

(b)

(c)