Improving bionic vision with deep learning

What is it about?

Bionic vision systems can help restore sight to those with permanent blindness. However, these systems are very limited and provide a highly distorted, often unrecognizable image of the user's surroundings. We present a deep learning encoder that, provided the target image, computes the ideal bionic vision system stimuli to produce artificial vision that closest matches the target image. We use a validated computational model of bionic vision and display the effectiveness of our method on data generated from three current users of bionic vision systems.

Featured Image

Photo by Marc Schulte on Unsplash

Photo by Marc Schulte on Unsplash

Why is it important?

Our 5 senses are the way we interact with the world, and some say sight is the most important. Unfortunately, millions of people in the United States alone suffer from debilitating blindness that hinders their everyday life. As current methods of sight restoration are elementary and not effective, finding a solution is critical to helping these patients earn their life back. We are the first paper to use a psycho-physically validated bionic vision model and evaluate our method on data based on actual users of prosthetic vision. Our work is an important first step in understanding the issues with current methods and discovering potential avenues for further exploration.