Semantic translation of face image with limited pixels for simulated prosthetic vision