Learned Block-Based Hybrid Image Compression

Recent works on learned image compression perform encoding and decoding processes in a full-resolution manner, resulting two problems when deployed for practical applications. First, parallel acceleration of the autoregressive entropy model cannot be achieved due to serial decoding. Second, inference often causes out-of-memory (OOM) problem with limited GPU resources, especially high-resolution images. Block partition is good choice handle above issues, but it brings about new challenges reducing redundancy between blocks eliminating block effects. To tackle challenges, this paper provides block-based hybrid (LBHIC) framework. Specifically, we introduce explicit intra prediction into framework utilize relation among adjacent blocks. Superior context modeling by linear weighting neighbor pixels traditional codecs, propose contextual module (CPM) better capture long-range correlations utilizing strip pooling extract most relevant information neighboring latent space, thus achieving effective prediction. Moreover, alleviate blocking artifacts, further boundary-aware postprocessing (BPM) edge importance taken account. Extensive experiments demonstrate that proposed LBHIC codec outperforms VVC, bit-rate conservation 4.1%, reduces time approximately 86.7% compared state-of-the-art methods.