Reading Pixelized Paragraphs of Chinese Characters Using Simulated Prosthetic Vision

Ying Zhao, Yanyu Lü, Zhao Ji, Kaihu Wang, Qiushi Ren, Kaijie Wu, Xinyu ChaiInvestigative 2011 · DOI

Abstract

Character size and resolution, pixel dropout, and the number of gray levels clearly affected the reading performance of pixelized Chinese paragraphs. Compared with pixelized character recognition, pixelized Chinese paragraph reading achieved higher accuracy; thus, optimal Chinese reading performance may require prostheses with more electrodes (1000) than are required to read paragraphs in the Latin alphabet (500).

Introduction

Visual prostheses, such as retinal or cortical implants, aim to restore functional vision by stimulating neurons to create "phosphenes"—discrete spots of light that form a pixelated image. While significant research has addressed reading requirements for Latin-based alphabets, Chinese characters present a unique challenge due to their high ideographic complexity. Unlike the 26 letters of the English alphabet, Chinese characters often consist of numerous intricate strokes. This study investigates the specific visual parameters—character size, resolution, pixel dropout, and gray-scale levels—necessary for successful Chinese paragraph reading.

Methods

The researchers conducted four experiments involving 40 native Chinese speakers using a simulated prosthetic vision system. Participants viewed pixelized Chinese paragraphs through a Head-Mounted Display (HMD).

Experimental design flowchart illustrating the manipulation of character resolution, size, dropout, luminance, and gray levels.

The simulation utilized circularly symmetric Gaussian distributions to model phosphenes, mimicking the stimulation patterns of real implants. The study systematically varied:

  • Resolution: 6×6, 8×8, and 10×10 pixels per character.
  • Character Size: Visual angles ranging from 1° × 1° to 9° × 9°.
  • Pixel Dropout: 0% to 50% "dead" pixels.
  • Luminance and Gray-scale: Varying levels of brightness and intensity steps.

The dataset comprised 200 Chinese paragraphs from primary school textbooks, using characters that represent approximately 92% of daily written information.

Results and Analysis

The study revealed several critical insights into the requirements for functional Chinese reading.

Resolution and Electrode Requirements

At a resolution of 10×10 pixels per character, subjects achieved nearly 100% reading accuracy. However, accuracy dropped to approximately 60% at 6×6 pixels. To maintain a functional reading speed of roughly 70 characters per minute with 90% accuracy, the authors estimate a prosthesis would require a resolution equivalent to 10×10 pixels per character, totaling approximately 1,000 electrodes across the functional field of view.

The Impact of Pixel Dropout

Chinese text proved highly sensitive to pixel dropout. While English readers can often infer words with significant missing data, a 50% dropout rate rendered Chinese paragraphs completely unreadable.

Visual comparison of Chinese characters across different resolutions (6x6 to 10x10) and pixel dropout percentages (0% to 50%).

Contextual and Spatial Factors

A significant "context bonus" was observed: reading accuracy was consistently higher for full paragraphs than for isolated characters, suggesting that top-down linguistic processing helps compensate for degraded image quality. Additionally, an optimal character size of 5° × 5° was identified. Deviations from this size resulted in either insufficient resolution (too small) or an exceeded perceptual span (too large), requiring excessive head scanning.

Discussion and Limitations

While these results provide a benchmark for prosthetic design, the study's reliance on normally sighted subjects and idealized Gaussian phosphenes is a limitation. Real-world prosthetic vision often involves irregular phosphene shapes and overlapping stimulation. Future work should explore dynamic text presentation and advanced image-processing algorithms to simplify complex characters before neural stimulation.

Summary updated

Abstract

Character size and resolution, pixel dropout, and the number of gray levels clearly affected the reading performance of pixelized Chinese paragraphs. Compared with pixelized character recognition, pixelized Chinese paragraph reading achieved higher accuracy; thus, optimal Chinese reading performance may require prostheses with more electrodes (1000) than are required to read paragraphs in the Latin alphabet (500).
BibTeX
@inproceedings{zhao2011reading,
  author = {Ying Zhao and Yanyu Lü and Zhao Ji and Kaihu Wang and Qiushi Ren and Kaijie Wu and Xinyu Chai},
  title = {{Reading Pixelized Paragraphs of Chinese Characters Using Simulated Prosthetic Vision}},
  booktitle = {Investigative},
  year = {2011},
  doi = {10.1167/iovs.10-5293},
}

Loading related content...

Figure