Neural Image Compression with Quantization Rectifier
This work addresses a specific bottleneck in neural image compression for applications requiring high-quality image reconstruction, representing an incremental improvement over existing methods.
The paper tackles the problem of quantization errors degrading image quality in neural image compression by introducing a quantization rectifier method that uses image feature correlation to predict unquantized features, resulting in consistent coding efficiency improvements on the Kodak benchmark with negligible runtime increase.
Neural image compression has been shown to outperform traditional image codecs in terms of rate-distortion performance. However, quantization introduces errors in the compression process, which can degrade the quality of the compressed image. Existing approaches address the train-test mismatch problem incurred during quantization, the random impact of quantization on the expressiveness of image features is still unsolved. This paper presents a novel quantization rectifier (QR) method for image compression that leverages image feature correlation to mitigate the impact of quantization. Our method designs a neural network architecture that predicts unquantized features from the quantized ones, preserving feature expressiveness for better image reconstruction quality. We develop a soft-to-predictive training technique to integrate QR into existing neural image codecs. In evaluation, we integrate QR into state-of-the-art neural image codecs and compare enhanced models and baselines on the widely-used Kodak benchmark. The results show consistent coding efficiency improvement by QR with a negligible increase in the running time.