Visual Attention Modeling for Autism Spectrum Disorder by Semantic Features

Autism spectrum disorder (ASD) is a common mental illnesses for children. Existing studies show that visual attention of children with ASD is different from that for normal children. Thus, it is meaningful to design an effective visual attention model for children to diagnose ASD. In this paper, we propose a saliency detection model for children with ASD by small-scale image samples. In the proposed model, we design a deep neural network by high-level semantic features to saliency detection in data-driven way. The U-net is used for construct the deep neural network for semantic feature learning, where a loss function of positive and negative equilibrium mean square-error (PN-MSE) is designed in the constructed deep neural network. Experimental results show that the proposed model can obtain better performance than other existing models.