Industrial Technology Research Institute

Shu-Hao Ye

Chi-Tien Sun

In a training phase of a facial image reconstruction system, a thermal facial image is input into a convolutional encoder and a pre-trained thermal facial recognition model for feature fusion. The fusion result is input into a convolutional decoder for visible facial image reconstruction. A hybrid loss function is used in the training phase. In the training phase, the reconstructed visible facial image and its corresponding real visible facial image are input into the pre-trained visible facial recognition model and the pre-trained facial landmark detection model to calculate the hybrid loss function for model learning and parameter update.

Growing community of inventors

Chiayi, Taiwan

Shu-Hao Ye