Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Deep Neural Networks for No-Reference and Full-Reference Image Quality Assessment.

Sebastian Bosse, Dominique Maniry, Klaus-Robert Muller

    IEEE Transactions on Image Processing : a Publication of the IEEE Signal Processing Society
    |October 14, 2017
    PubMed
    Summary

    A novel deep neural network achieves superior image quality assessment (IQA) in both no-reference (NR) and full-reference (FR) settings. This data-driven approach learns local quality and importance for robust, generalizable results across diverse image databases.

    Related Concept Videos

    You might also read

    Related Articles

    Articles linked to this work by shared authors, journal, and citation graph.

    Sort by
    Same author

    Structural and mechanistic insights into primer synthesis initiation by DNA primase.

    Nature communications·2026
    Same author

    Towards robust foundation models for digital pathology.

    Nature communications·2026
    Same author

    Beyond attention heatmaps: How to get better explanations for multiple instance learning models in histopathology.

    Medical image analysis·2026
    Same author

    Spotlight on the Nucleotide: Solid-State NMR for the Investigation of ATP Hydrolysis in the ATPase SmsC.

    Chemistry (Weinheim an der Bergstrasse, Germany)·2026
    Same author

    Controlling the Guanidinium Cation Rotation by Cation-π Interactions.

    Angewandte Chemie (International ed. in English)·2026
    Same author

    AI-based discovery of functional boundaries in the human brain from intraoperative electrophysiology.

    medRxiv : the preprint server for health sciences·2026

    Area of Science:

    • Computer Vision
    • Machine Learning
    • Image Processing

    Background:

    • Image quality assessment (IQA) is crucial for evaluating digital images.
    • Existing IQA methods often rely on hand-crafted features or specific domain knowledge.
    • Deep learning offers potential for more adaptive and accurate IQA.

    Purpose of the Study:

    • To develop a deep neural network (DNN) for advanced image quality assessment.
    • To create a unified framework for both no-reference (NR) and full-reference (FR) IQA.
    • To demonstrate the data-driven superiority of the proposed DNN over existing methods.

    Main Methods:

    • An end-to-end trained deep neural network with ten convolutional and five pooling layers for feature extraction.
    • Two fully connected layers for regression, enabling joint learning of local quality and weights.

    Related Experiment Videos

  • A purely data-driven approach, avoiding reliance on human visual system models or image statistics.

    • Main Results:

    • The proposed DNN achieved superior performance compared to state-of-the-art NR and FR IQA methods on multiple benchmark databases (LIVE, CISQ, TID2013, LIVE In the wild).
    • The architecture demonstrated flexibility, allowing adaptation for both NR and FR IQA settings.
    • Cross-database evaluation confirmed high generalization ability and robustness of the learned features.

    Conclusions:

    • The developed deep neural network provides a highly effective and robust solution for image quality assessment.
    • The unified framework and data-driven approach offer significant advantages over traditional IQA methods.
    • The model's strong cross-database generalization highlights its potential for real-world applications.