Cross-Layer Hybrid Compression for Efficient Fundus Image Transmission in Machine Vision Systems
Yingchun Zhang
Hangzhou Rangji Technology Co., Ltd., Hangzhou, Zhejiang Province, 310030, China.
Jiangke Wu *
School of Informatics, Computing and Cyber Systems Northern Arizona University, Flagstaff, Arizona, 86011, United States.
*Author to whom correspondence should be addressed.
Abstract
To address the trade-off between bandwidth bottlenecks and feature fidelity in RESTful image transmission for tele-ophthalmology IoT systems, we propose a machine-vision-oriented cross-layer collaborative compression framework. This scheme integrates lossy dimensionality reduction (JPEG) at the semantic layer with entropy coding optimization (Zlib) at the transport layer. Utilizing the Messidor-2 dataset, we evaluated the feature consistency of ResNet-18 and MobileNetV2. The experiments identified an ”Empirical Inflection Point” at Q = 0.4, where data volume was reduced by 57.3% while feature space cosine similarity remained above 0.920. Although classification accuracy experienced a minor 4% decline, the system maintained high inference stability. Furthermore, the Zlib strategy successfully eliminated the 33% encoding redundancy introduced by the Base64 protocol. This framework demonstrates that by co-optimizing the semantic and transport layers, low-latency transmission can be balanced with AI reliability, offering a scalable reference for broader medical IoT systems operating in bandwidth-constrained environments.
Keywords: Tele-ophthalmology, image compression, feature consistency, RESTful API, cross-layer