Inclusive Convolutional Neural Network Design Enabling Partially Sighted People to Expand Viewing-Experience on Smart Screens

A deep neural network-based picture enhancement technique that enables partially sighted people to expand their viewing-experience on smart large TV screens is proposed. Reflecting insights from our previous studies on preferred picture enhancement features for low vision people, a convolutional neural network architecture that can generate visibility-enhanced images on screen is presented. The neural network which has very large scales of convolutioinal layers is trained to output super-resolved and salient feature-improved images for helping the visually impaired to see more clearly images on screens. Our experiment result proves that synthesized images by the proposed neural network are expected to give more vivid visual experiences when people with low vision are watching screens. To the best of our knowledge, inclusive neural network design in terms of the picture quality is the first approach which can help the visually impaired to see directly any content itself on screen.