Researchers from NESTOR Project 2 have published their latest findings on the benefits and feasibility of using software algorithms to process and generate artificial vision. Their work, by de Ruyter van Steveninck et al., published in the Journal of Vision, describes their results using contour-based scene simplification at different phosphene resolutions in a simulation experiment on sighted participants.
They explored both the theoretically attainable benefits of strict scene simplification in an indoor environment (by controlling the complexity of the surrounding environment), as well as the practical results attained using two methods: a deep-learning-based surface boundary detection implementation, and traditional edge detection.
They found that scene simplification requires a careful tradeoff between informativeness and interpretability of artificially generated visual percepts, which may depend on whether a larger or smaller number of implanted electrodes are used.