COGL: Coefficient Graph Laplacians for Optimized JPEG Image Decoding.

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Decoding Joint Photographic Experts Group (JPEG)-encoded images often results in visual artifacts.
• Existing methods for image decoding can be computationally intensive and complex.

Purpose of the Study:

• To develop an efficient and effective method for decoding JPEG-encoded images with reduced visual artifacts.
• To propose a novel graph-based regularization model for image decoding.

Main Methods:

• The decoding task is framed as an ill-posed inverse problem solved using a convex, graph Laplacian-regularized model.
• Fast high-dimensional Gaussian filtering and proximal gradient descent are employed for efficient convex problem solving.
• A patch-based coefficient graph is utilized for regularization, outperforming traditional pixel-based approaches for natural images.

Main Results:

• The proposed method achieves decoded image quality comparable to state-of-the-art techniques.
• The method demonstrates a significant speed improvement, being up to five times faster than existing complex methods.
• The approach is extended to handle Motion JPEG (M-JPEG)-encoded video decoding.

Conclusions:

• The convex, graph-regularized model provides an efficient and effective solution for JPEG image decoding.
• The patch-based graph regularization is well-suited for non-stationary signals like natural images.
• The method offers a faster alternative to complex decoding techniques without compromising quality.