Single image sub-pixel rendering using a residual network for flat panel display

Area of Science:

• Display technology
• Image processing
• Computer vision

Background:

• Increasing pixel density in flat panel displays, particularly Micro-LEDs, raises manufacturing costs and complexity.
• Subpixel rendering is crucial for high-resolution displays but faces challenges with color accuracy and luminance resolution.

Purpose of the Study:

• To propose a novel single-image subpixel rendering method for RGBG square-arranged flat panel displays.
• To address the trade-off between luminance resolution and chromatic aliasing in subpixel rendering.
• To reduce the manufacturing difficulty and cost associated with high-density displays.

Main Methods:

• A residual network is employed for nonlinear mapping from the original image to the rendered subpixel image.
• A perceptual error loss function, based on the human eye's contrast sensitivity function (CSF), is designed.
• The method is specifically tailored for RGBG square-arranged subpixel displays.

Main Results:

• The proposed method demonstrates significant advantages in rendering subpixel images.
• Effective control of color errors is achieved while maintaining image clarity.
• The method outperforms existing subpixel rendering algorithms on key metrics like zoomed peak-signal-to-noise-ratio (PSNRz) and perceptual error measuring (PEM).

Conclusions:

• The novel subpixel rendering method offers a viable solution to challenges in high-density display manufacturing.
• The approach effectively balances luminance resolution and chromatic aliasing, enhancing visual quality.
• The method shows significant potential for improving the visual fidelity of Micro-LED and other advanced flat panel displays.