Perception Enhancement For Bionic Vision Preliminary Study On Object Classification With Subretinal Implants

The restored vision by using subretinal implants of patients suffering from a loss of photoreceptors, e.g., in retinitis pigmentosa and age-related macular degeneration, is, compared to healthy subjects, very limited. Therefore, we investigated, whether it is possible to enhance the perception of such patients by transforming the input images in a systematic manner. To this end, we propose a new image transformation network that is capable to learn plausible image transformations in an end-to-end fashion in order to enhance the perception of (virtual) patients with simulated subretinal implants. As a proof of concept, we test our method on an object classification task with three classes. Our results are promising. Compared to a baseline model, the overall object classification accuracy increased significantly from 67.4% to 81.1%. Furthermore, we discuss implications and limitations of our proof of concept and outline aspects of our work that can be improved and need to be subject of further research.