JPEG quality transcoding using neural networks trained with a perceptual error measure

¹CS Division, University of California, Berkeley, 413 Soda Hall, Berkeley CA 94720, USA. lazzaro@cs.berkeley.edu

|February 9, 1999

Summary

This study introduces a JPEG Quality Transcoder (JQT) that reduces visual artifacts in low-quality JPEG images without the original file. The pattern recognition technology successfully removes over 30% of compression artifacts, enhancing image quality.

Area of Science:

Computer Vision
Image Processing
Digital Signal Processing

Background:

JPEG compression introduces visual artifacts, degrading image quality.
Restoring image quality often requires the original uncompressed image, which is typically unavailable.
Existing methods struggle to effectively reduce artifacts in transcoded JPEG images.

Purpose of the Study:

To develop a JPEG Quality Transcoder (JQT) technology for artifact reduction in compressed images.
To design a JQT using a pattern recognition approach and statistical models.
To incorporate human visual perception models for artifact error measurement.

Main Methods:

Utilizing a pattern recognition approach for JPEG Quality Transcoder (JQT) design.
Training statistical models on a database of images to learn JPEG compression artifacts.
Employing a human visual perception model as an error metric during model training.

Main Results:

The prototype JPEG Quality Transcoder (JQT) system demonstrated artifact removal capabilities.
Achieved removal of 32.2% of artifacts from moderately compressed images.
Measured artifact reduction on an independent test database using a perceptual error metric.

Conclusions:

The developed JPEG Quality Transcoder (JQT) technology effectively reduces visual artifacts in JPEG images.
Pattern recognition and statistical modeling offer a viable approach for JPEG artifact reduction without original data.
The system shows promise for improving the visual quality of transcoded JPEG images.

Related Experiment Videos

JPEG quality transcoding using neural networks trained with a perceptual error measure

Related Concept Videos

Atropisomeric quinazolin-4-one derivatives are potent noncompetitive alpha-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor antagonists.

Designing a connectionist network supercomputer.

The design of a neuro-microprocessor.

Silicon auditory processors as computer peripherals.

Analog VLSI model of binaural hearing.

Silicon modeling of pitch perception.

Shape Matters: Few-Shot Object Classification From High-Information Contour Features.

Associative Memory and Generative Diffusion in the Zero-Noise Limit.

Neuronal Spike Trains as Functional-Analytic Distributions: Representation, Analysis, and Significance.

Quantifying Information Stored in Synaptic Connections Rather Than in Firing Activities of Neural Networks.

Unsupervised Feature Selection Using Bayesian Tucker Decomposition.

Cross-Frequency Coupling as a Neural Substrate for Prediction Error Evaluation: A Laminar Neural Mass Modeling Approach.

Evaluation and application of automated quality control in clinical coagulation testing.

Validation of expected objective performance indicators in surgical data analysis.

Internal feasibility evaluation of SAM-Med3D for cervical cancer clinical target volume segmentation on planning computed tomography.

Optimizing Hip and Knee Arthroplasty Clinic Flow: A Prospective Evaluation of Artificial Intelligence Scribe Technology.

Automated Semisupervised Measurement of Optic Nerve Sheath Diameter From CT Following Traumatic Brain Injury.

Association of Asynchronous Virtual Scribe Services: Patient Experience in Primary Care.

Related Experiment Videos

JPEG quality transcoding using neural networks trained with a perceptual error measure

Area of Science:

Background:

Purpose of the Study:

Main Methods:

Main Results:

Conclusions:

Related Concept Videos

Related Articles