A Model-Driven Deep Unfolding Method for JPEG Artifacts Removal.

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Lossy JPEG compression introduces blocking artifacts, degrading image quality.
• Current deep learning methods for artifact removal lack interpretability, hindering performance improvements.

Purpose of the Study:

• To develop a model-driven deep unfolding method for JPEG artifact removal.
• To create an interpretable network structure for enhanced deblocking performance.

Main Methods:

• Constructed a maximum posterior (MAP) model for deblocking using convolutional dictionary learning.
• Designed an iterative optimization algorithm with proximal operators.
• Unfolded the iterative algorithm into a learnable deep network structure for end-to-end training.

Main Results:

• The proposed method achieves competitive or superior deblocking results compared to state-of-the-art techniques.
• Experiments on synthetic and real-world datasets validate the effectiveness of the approach.
• The network effectively characterizes JPEG artifacts and image content representations.

Conclusions:

• The model-driven deep unfolding method offers both powerful performance and interpretability in JPEG artifact removal.
• This approach combines the strengths of data-driven deep learning and traditional model-driven techniques.
• The interpretable network structure facilitates further advancements in image deblocking technology.