Scene Simplification for Simulated Prosthetic Vision with Improved Scene Understanding.

Visual impairment or blindness affects millions of people worldwide, causing significant challenges in their daily activities and communication. These people are potentially a special group of users in future consumer electronics applications. Retinal prostheses have emerged as a promising solution for restoring vision in individuals with visual impairments. These prostheses can compute and display simplified representations of images to visually impaired individuals. Therefore, prostheses can be considered special consumer electronics devices. However, the quality of the generated images by these prostheses remains unsatisfactory. Moreover, advanced computer vision algorithms have been employed for image processing to enhance prosthetic vision’s applicability. Recent studies have utilized object segmentation for scene simplification, but insufficient structural information compromises scene understanding in object-absent scenes. To tackle this challenge, in our study, we have utilized an image fusion technique to integrate saliency information based on the semantic segmentation results of previous work. These fused images can improve the scene understanding of visually impaired individuals. To verify the effectiveness of our approach, a user study was conducted. The user study results, including the sensitivity coefficient and other relevant indicators, demonstrated notable improvements. We believe that this work will contribute to enhancing information accessibility for visually impaired individuals, thereby aiding their overall well-being.