Gaze-enhanced genetic optimization for near-eye display video compression

With the rapid advancement of immersive technologies, near-eye display devices have established themselves as fundamental interfaces for next-generation human-computer interaction systems, revolutionizing how visual information is presented and consumed. The exponential growth of video data in these applications has created pressing challenges in both real-time data transmission and storage efficiency, demanding novel coding approaches that are specifically optimized for human visual perception characteristics. While modern deep learning-based video coding architectures have demonstrated remarkable success in reducing spatiotemporal redundancy, their conventional implementation often adopts a homogeneous encoding paradigm that fails to account for the inherent non-uniformity of human visual perception, resulting in suboptimal performance in adaptively encoding visually salient regions where perceptual quality matters most. To bridge this critical gap, we present a novel intelligent video compression framework that synergistically combines gaze perception analytics with advanced genetic optimization techniques. Our work makes two primary theoretical and practical contributions to the field: Gaze-Aware Dynamic Compression Ratio Optimization Algorithm for Single Frame Images：By integrating binocular gaze perception data with multi-scale spatial pyramid decomposition, we construct a Bayesian probability-based visual saliency weighting model. The implementation consists of three key steps: First, multi-scale feature extraction: Laplacian pyramid transform is employed to extract visual features across different resolutions, obtaining spatio-frequency joint characteristics. Second, saliency fusion: Through probability density coupling between fixation heatmaps and computational saliency maps, a pixel-wise visual importance distribution is generated. Finally, adaptive quantization: Based on the saliency weights, a non-uniform quantization matrix is derived to achieve dynamic bit allocation under a rate-distortion optimization framework, thereby enhancing compression efficiency while preserving visual quality. Evolutionary computing-enhanced saliency prediction for temporal sequences: Extending our single-frame optimization principles, we devise an innovative spatiotemporal analysis framework powered by genetic algorithm optimization. This advanced model establishes complex nonlinear correlations between inter-frame motion dynamics and visual attention patterns, enabling the creation of an intelligent cross-pyramid saliency propagation network. The system employs genetic algorithms to autonomously discover and refine the intricate relationships between time-varying visual features and gaze-directed regions of interest. Furthermore, we implement a multi-resolution spatiotemporal fusion architecture that precisely models and predicts the propagation characteristics of salient regions across different pyramid hierarchy levels. The culmination of this research is an intelligent hierarchical video encoding system that leverages these predictive models to achieve unprecedented coding efficiency gains while maintaining superior perceptual quality standards.