2.5. Conclusion 2 23 Sanchez-Garcia et al., 2020; Vergnieux et al., 2017), a limitation of our study is that we did not include this condition in our experiments. Including greyscale intensity mapping as a study condition would have enabled us to compare the effect of contour-based scene simplification to a less-restricted control condition with SPV. 2.5. Conclusion Investigating suitable computer vision strategies for scene simplification is an important step in the development of visual prostheses. Our results suggest that contour-based simulated prosthetic vision with a resolution of 26 × 26 phosphenes provides adequate information for mobility. Strict scene simplification with surface boundary extraction may help to overcome visual overcrowding at lower phosphene resolutions. However, the presence of within-surface information and background textures improves performance at higher phosphene resolutions. Therefore, choosing a balanced amount of information reduction is advised, depending on the number of implanted electrodes. Currently, the implementation of deep learning models for surface-boundary detection in a real-time mobility task remains challenging and future research and empirical validation is required to further explore the potential of this approach.
RkJQdWJsaXNoZXIy MTk4NDMw