Object recognition from sparse simulated phosphenes and curved segments

Cortical prostheses offer the potential for partial vision restoration in individuals with blindness by stimulating neurons to produce phosphenes. However, the low number of discrete phosphenes that can be simultaneously elicited in practice makes encoding of whole objects difficult, hindering recognition. Hence, we aim at determining the minimal number of discrete visual elements needed for object recognition. To answer this question, we replaced object contours by a varying number of round patches, serving as idealized phosphene simulation. Forty-six sighted participants identified the fragmented objects in a free-naming task. We increased the number as well as the patch density gradually until the objects were correctly identified. Depending on the object, we found that a minimum of 29 and a maximum of 65 round patches were necessary for recognition. While this still exceeds the current number of phosphenes which can simultaneously be elicited in cortical visual protheses, we found much smaller estimates than previous studies of simulated prosthetic vision. We next investigated whether local visual information of non-circular patches could be leveraged to reduce the number of discrete elements necessary for object recognition. For this, objects were fragmented into straight and curved segments matching the size, density, and location of the previously used round patches. Participants required 27% fewer segments than round patches for recognition, indicating that contour information leads to perceptual benefits. Our study provides a lower bound estimate on the minimal number of phosphenes needed to recognize objects, should simultaneous stimulation protocols be pursued. These findings provide realistic bounds for the design of visual prostheses.