Article
Beyond downsampling: semantic preservation in retinal implant stimuli
Search Medline for
Authors
-
Henning Konermann
- Institute of Imaging and Computer Vision, RWTH Aachen University, Aachen, Germany
-
Y. Wu
- Institute of Imaging and Computer Vision, RWTH Aachen University, Aachen, Germany
-
P. Walter
- Department of Ophthalmology, RWTH Aachen University, Aachen, Germany
-
J. Stegmaier
- Institute of Imaging and Computer Vision, RWTH Aachen University, Aachen, Germany
| Published: | May 9, 2025 |
|---|
Outline
Text
Objective: We propose a novel pipeline for image information reduction, designed to preserve semantic content when used with a stimulus encoder for retinal implants.
Materials and Methods: The proposed pipeline aims to address the limitations of current retinal implants, including low resolution and visual distortions. The approach includes saliency-based zooming, segmentation-driven scene simplification, and feature shift-invariant non-uniform downsampling. This method is designed to work alongside a deep learning-based stimulus encoder to optimize the visual stimuli for low-resolution implant displays.
Results: The expected outcome of the proposed method is to maintain crucial visual information despite image downsampling. By preserving semantic content, the pipeline seeks to improve the quality of the visual stimuli presented to patients with retinal implants.
Discussion: This approach aims to overcome the limitations of conventional downsampling methods, which often fail to retain semantically essential features in reduced-resolution images. If successful, this method could enhance the visual experience for retinal implant users by providing clearer, more meaningful stimuli.
Acknowledgment: This work was supported by Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) with the grant GRK2610: InnoRetVision (project number 424556709).
