Conditional Encoder-Based Adaptive Deep Image Compression with Classification-Driven Semantic Awareness

This paper proposes a new algorithm for adaptive deep image compression (DIC) that can compress images different purposes or contexts at rates. The with semantic awareness, which means classification-related features are better protected in lossy compression. It builds on the existing conditional encoder-based DIC method and adds two features: model-based rate-distortion-classification-perception (RDCP) framework to control trade-off between rate performance contexts, mechanism generate coding conditions based complexity importance. outperforms QMAP2021 benchmark ImageNet dataset. Over tested range, it improves classification accuracy by 11% perceptual quality 12.4%, 32%, 1.3% average NIQE, LPIPS, FSIM metrics, respectively.