Foveated image compression with block-based optimization for VR

With advances in metaverse technologies, semiconductor processes, and artificial intelligence, near-eye displays (AR/VR) have become more miniaturized, integrated, and intelligent. Microdisplays, as the core imaging component, determine visual experience. OLED on silicon are favored for their self-emission, high contrast, fast response, and high pixel density. However, rising demands for resolution, refresh rate, and grayscale exceed analog driving capabilities, while digital driving increases data transmission, creating a bandwidth bottleneck. This study proposes a foveated image compression method that leverages human visual gaze point and the Foveated Just-Noticeable Distortion (FJND) model. Gaze point preprocessing combined with FJND removes visually imperceptible information. Bilinear interpolation reduces hardware complexity, and multi-layer scaling adjusts resolution region-wise. The FJND processing removes visually redundant information that is imperceptible to the human eye. A hardware-friendly pixel-block encoding scheme, optimized via a genetic algorithm, achieves significant compression (overall rate 20.25%) while preserving perceptual quality. FPGA validation confirms stable application on 2560×2560×3 OLED microdisplays with no noticeable difference in the gaze region, providing a practical perception-driven compression solution for VR/AR displays.