Abstract
The transition to 6G has intensified the need for semantic-aware communication, surpassing Shannon's bit-level limits to meet intelligent application demands. However, visual semantic communication lacks a unified theory, has inconsistent definitions of semantic similarity, and suffers from fragmented designs. This paper formulates a Kantian epistemology-inspired framework-Unified Hierarchical Epistemological Semantic Communication (ESC). It operationalizes the three-tier cognitive schema of Sensibility, Understanding, and Reason as both a system design principle and a multi-dimensional evaluation metric for human-centric image transmission. To validate this framework, we propose the Knowledge-Infer-Semantic-Sketch (KISS) scheme, a dual-branch system developed according to the proposed methodology. The geometric branch uses contour extraction, semantic filling, and graph-based reasoning for decoupled scenarios, while the vector branch employs a hierarchical VQ-VAE for coupled scenarios. Novel algorithms enable semantically sparse representation, noise-resilient constellation mapping, and semantic-tampering error correction. Evaluated on benchmarks such as Cityscapes and COCO, KISS achieves superior performance across three-dimensional metrics: visual details (SSIM, LPIPS), categorical understanding (mIoU), and semantic reasoning (CLIP)-demonstrating comprehensive gains in compression, interpretability, and adaptability. This framework provides an extensible, human-centric foundation that redefines visual semantic communication for 6G and beyond.