Enhancing E-Paper Color Fidelity: A Deep Learning Approach to Color Correction

Color electronic paper (e-paper) with a printed color filter array (CFA) offers fast response times and ultra-low power consumption, making it a promising display technology. However, its color gamut is notably narrower than that of multipigment e-paper or liquid crystal displays (LCDs) due to reduced grayscale levels-16 per pixel in CFA-based e-paper compared to 256 levels per RGB component in LCDs-posing challenges for accurate color reproduction. While dithering techniques, such as error diffusion, have been used to improve image quality, their high computational cost (e.g., 75 seconds per image) conflicts with the fast response characteristics of CFA-based e-paper. This study proposes a deep learning-based color correction framework specifically designed for e-paper. Using depthwise separable convolutional neural networks, the framework achieves efficient and precise color adjustments, with a peak signal-to-noise ratio (PSNR) of 24.33 dB and a structural similarity index measure (SSIM) of 0.92. Remarkably, it processes images in 1 second on standard hardware, delivering a 75-fold speedup over traditional dithering methods. Experimental results demonstrate improved color fidelity, visual quality, and generalizability, positioning the framework as a transformative solution to the color reproduction challenges of CFA-based e-paper.