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

Multiscale hybrid linear models for lossy image representation.

Wei Hong1, John Wright, Kun Huang

  • 1DSP Solutions Research and Development Center, Texas Instruments, Dallas, TX 75243, USA. weihong@ti.com

IEEE Transactions on Image Processing : a Publication of the IEEE Signal Processing Society
|December 13, 2006
PubMed
Summary
This summary is machine-generated.

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

Vulnerabilities in child well-being among primary school children: a cross-sectional study in Bradford, UK.

BMJ open·2022
Same author

mRNA or ChAd0x1 COVID-19 Vaccination of Adolescents Induces Robust Antibody and Cellular Responses With Continued Recognition of Omicron Following mRNA-1273.

Frontiers in immunology·2022
Same author

Urban environment and health behaviours in children from six European countries.

Environment international·2022
Same author

Study of the Combined Effect of Maternal Tobacco Smoking and Polygenic Risk Scores on Birth Weight and Body Mass Index in Childhood.

Frontiers in genetics·2022
Same author

Integrating polygenic risk scores in the prediction of type 2 diabetes risk and subtypes in British Pakistanis and Bangladeshis: A population-based cohort study.

PLoS medicine·2022
Same author

The early-life exposome modulates the effect of polymorphic inversions on DNA methylation.

Communications biology·2022

This study introduces a multiscale hybrid linear model for efficient natural image representation. This novel approach, using generalized principal component analysis (GPCA), yields more compact image data compared to existing methods.

Area of Science:

  • Computer Vision
  • Image Processing
  • Machine Learning

Background:

  • Existing image representation methods often rely on fixed or uni-modal transformations, limiting their ability to capture complex data correlations.
  • These methods include Discrete Cosine Transform (DCT), wavelets, Principal Component Analysis (PCA), and Vector Quantization (VQ).

Purpose of the Study:

  • To introduce a simple and efficient representation for natural images.
  • To develop a novel multiscale hybrid linear model that exploits multimodal correlations across different scales.
  • To address limitations of current image representation techniques.

Main Methods:

  • Representing images as collections of vectors in high-dimensional space.
  • Fitting a piece-wise linear model (union of affine subspaces) at each downsampling scale.

Related Experiment Videos

  • Estimating the model using generalized principal component analysis (GPCA).

    • Main Results:

    • The multiscale hybrid linear model effectively extracts and utilizes multimodal correlations in image data.
    • The model reduces both complexity and computational cost compared to existing methods.
    • Experimental results demonstrate more compact representations for natural images across various signal-to-noise ratios.

    Conclusions:

    • The proposed multiscale hybrid linear model offers a significant improvement in image representation efficiency.
    • GPCA provides an effective algebraic method for estimating this hybrid model.
    • The modeling paradigm shows potential for extensions to other applications like image segmentation.