Block-based Learned Image Compression without Blocking Artifacts

Learned Image Compression (LIC) outperforms traditional codecs but suffers from excessive peak memory usage when handling high-resolution images. Consequently, blockbased LIC has been studied to reduce peak memory and peak computational cost but often introduces blocking artifacts that degrade visual quality. To mitigate this, the JPEG-AI standard introduced a patch-based scheme where overlapped blocks are coded independently using empirically determined overlap sizes. However, the experimental search for optimal overlaps is time-consuming and does not guarantee blocking-free reconstruction.

In this paper, we propose an analytic framework modeling overlap propagation through convolution and transposed convolution layers to precisely determine the minimal overlaps for blocking-free reconstruction. Based on the calculated minimum overlaps, we provide the block-based implementation methodology that could be applied to most CNN-based LIC models. Applied to four CNN-based LIC models on 4K images partitioned into various block sizes (256×256, 512×512), our method achieves rate-distortion performance identical to full-image coding while reducing average peak memory usage to 13.94% (encoder) and 13.33% (decoder), and average peak computational cost to 2.6% and 1.24%, respectively. Notably, the proposed blockbased framework does not require any re-training of the original model. Furthermore, it can also be applied to most CNN-based image processing neural networks without worrying about any performance degradation.