Enhancing Facial Recognition in Visual Prostheses using Region of Interest Magnification and Caricaturing.

Blindness is a prevalent disability with significant personal and societal consequences. While medical advancements offer treatment options, severe damage to the retina, optic nerve, or brain may remain untreated. Visual prostheses are implantable medical devices that aim at providing limited vision to such individuals. However, such prostheses offer low spatial resolution making activities like reading, facial recognition, and navigation challenging. This work aims to enhance implantees’ ability to recognize faces through real-time scene preprocessing, machine learning and computer vision techniques. Virtual-reality visual models simulating prosthetic vision were tested on normally/corrected sighted subjects, investigating the use of histogram equalization for contrast enhancement, facial region magnification, and caricaturing of facial features. Results revealed that histogram equalization with magnification increases facial recognition accuracy by 60%, distinguishability accuracy by 50%, and accuracy of seeing facial details by 90%. In contrast, adding facial caricaturing improved the accuracies by 66.66%, 25%, and 75% for recognizability, distinguishability, and seeing facial details, respectively. Consequently, the combination of visual field histogram equalization, face magnification, and optional caricaturing can be considered as a promising enhancement approach that could enhance the quality of vision perceived through visual prostheses.