Text image processing for visual prostheses.

Area of Science:

• Biomedical Engineering
• Computer Vision
• Ophthalmology

Background:

• Retinal diseases cause significant vision loss globally.
• Visual prosthetics offer partial vision restoration.
• Reading is a crucial functional ability impacted by vision loss.

Purpose of the Study:

• To develop a text image processing strategy for visual prostheses.
• To enhance text readability in low-resolution phosphene vision.
• To simulate and evaluate the proposed strategy's effectiveness.

Main Methods:

• A BP neural network was used for text recognition from camera images.
• Recognized text was simplified into 5x7 pixel characters.
• A simulation model mimicked prosthetic vision with phosphene arrays.
• Phosphene letters were displayed and evaluated for clarity and readability.

Main Results:

• The proposed strategy produced more intact and clearer phosphene letter contours.
• Simulated prosthetic vision with the strategy allowed for readily readable phosphene letters.
• The processed phosphene letters were superior to those displayed without the strategy.

Conclusions:

• The developed text image processing strategy benefits reading for blind patients using visual prosthetics.
• This approach enhances the clarity and readability of text in phosphene vision.
• The strategy shows promise for improving the reading experience in prosthetic vision applications.