Neural Network Reduces JPEG Blocking and Ringing Artifacts

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

JPEG compression introduces artifacts such as blocking and ringing in decompressed images, particularly noticeable in low bit rate compression, which degrade image quality.

Innovation Solution

A neural network is trained to reduce artifacts in spatial domain representations of images compressed using transform methods like JPEG, by processing input vectors formed from pixel neighborhoods to predict and correct pixel values, thereby smoothing blocking and ringing artifacts.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of substance

If JPEG compression is performed at low bit rate to reduce file size, then data compression ratio is improved, but blocking and ringing artifacts increase degrading image quality

Engineering Contradiction:

Improvefile sizeVSAvoidblocking and ringing artifacts

Core Design Contradiction:

Loss of substanceVSObject-affected harmful factors

Solution Approach 1:

A neural network is introduced as an intermediary component between the JPEG decompression and the final image output. The neural network processes the decompressed image data and selectively reduces blocking and ringing artifacts while preserving important image features, thus mediating between the compressed input and the quality output without requiring re-compression

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The invention converts the harmful artifacts (blocking and ringing) into beneficial information by training the neural network to recognize these artifact patterns and distinguish them from genuine image features. The network learns from training pairs where artifacts are present to identify and suppress only the artifact components while preserving real image content

Inventive Principle:

Principle #22Blessing in disguise (Convert harm into benefit)

2Productivity

If quantization is applied to reduce coefficient precision, then compression efficiency is improved, but image reconstruction accuracy deteriorates

Engineering Contradiction:

Improvecompression efficiencyVSAvoidimage reconstruction accuracy

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The invention replaces the traditional mechanical/mathematical image reconstruction process (inverse DCT on quantized coefficients) with a neural network-based system. Instead of directly reconstructing from quantized coefficients, the neural network learns the mapping from compressed to high-quality images, substituting the rigid mathematical reconstruction with a flexible learned model that can compensate for quantization losses

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS7346208B2Image artifact reduction using a neural network

Publication Date: 2008.03.18 HEWLETT PACKARD DEVELOPMENT COMPANY LP

AI summary

A neural network is trained and used to reduce artifacts in spatial domain representations of images that were compressed by a transform method and then decompressed. For example, the neural network can be trained and used to reduce artifacts such as blocking and ringing artifacts in JPEG images.