Abstract:Digital images in real world applications typically undergo a wide variety of quality degradations before compression or re-compression. Existing learning based codecs are typically data-driven, relying on the predefined compression pipeline with pristine or high quality images as the input. However, the images in the wild may exhibit the substantially different characteristics compared to the high quality images, casting major challenges to the learning based image coding. In this paper, we propose a robust noisy image compression framework with the blind assumption on the specific noise type and level. The specifically designed encoder decomposes the representation of visual content into two types of features, including the Features that represent the Intrinsic Content (FIC) and the Features that account for Additive Degradation (FAD). As such, beyond the philosophy of faithfully reconstructing the given image with high fidelity, only FIC needs to be compactly represented and conveyed. The principled disentanglement strategy facilitates the removal of the redundancy from multiple perspectives (e.g., spatial, channel and content), ensuring the handling of a wide variety of noisy images in the wild. Extensive experimental results show that our model can achieve superior performance in terms of the ultimate quality and exhibit the strong generalizability across images degraded by a variety of means. The proposed scheme also points out a new research avenue on learning based compression for images in the wild, which is technically challenging but desirable in practice.

Low-Rank Decomposition-Based Restoration of Compressed Images via Adaptive Noise Estimation.

Quality Adaptive Low-Rank Based JPEG Decoding with Applications

Spatial-Adaptive Network for Single Image Denoising

A Novel Image Decompression Technique Using Image Restoration And Coefficient Fusion

Learning-based Compression for Noisy Images in the Wild

Learning-Based Image Restoration for Compressed Image through Neighboring Embedding

Near-lossless L-infinity Constrained Multi-rate Image Decompression Via Deep Neural Network

Compression Noise Estimation and Reduction Via Patch Clustering

Content-Adaptive Image Compressed Sensing Using Deep Learning

Near-lossless ` ∞-constrained Multi-rate Image Decompression via Deep Neural Network

Learning To Restore Low-Light Images Via Decomposition-And-Enhancement

Learning-based image restoration for compressed images

Weighted bilinear factorization of low-rank matrix with structural smoothness for image denoising

Sparsity-Promoting Adaptive Coding with Robust Empirical Mode Decomposition for Image Restoration.

Robust image restoration via adaptive low-rank approximation and joint kernel regression.

Image Denoising Via Structure-Constrained Low-Rank Approximation

From Rank Estimation to Rank Approximation: Rank Residual Constraint for Image Restoration

Low-Light Image Enhancement via Retinex-Style Decomposition of Denoised Deep Image Prior

Image Compression Algorithm Using Image Restoration Based on Wavelet Analysis

Image Denoising Via Low Rank Regularization Exploiting Intra and Inter Patch Correlation

Powerful Lossy Compression for Noisy Images