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

Fast vector quantization algorithms based on nearest partition set search.

Shen-En Qian1

  • 1Canadian Space Agency, Saint-Hubert, QC J3Y 8Y9, Canada. shen-en.qian@space.gc.ca

IEEE Transactions on Image Processing : a Publication of the IEEE Signal Processing Society
|August 12, 2006
PubMed
Summary

A new fast search method speeds up vector quantization by intelligently searching partitions in the generalized Lloyd algorithm (GLA). This significantly reduces computation time for codebook training while maintaining identical results.

Related Concept Videos

You might also read

Related Articles

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

Sort by
Same journal

Marginal-Aware Framework for 3D Shape Segmentation: Resolving Boundary-Internal Face Imbalance.

IEEE transactions on image processing : a publication of the IEEE Signal Processing Society·2026
Same journal

Semantic Composition via Optimal Transport for Composed Image Retrieval.

IEEE transactions on image processing : a publication of the IEEE Signal Processing Society·2026
Same journal

MultiTaskVIF: Segmentation-oriented visible and infrared image fusion via multi-task learning.

IEEE transactions on image processing : a publication of the IEEE Signal Processing Society·2026
Same journal

DTEA: Degradation-Aware Taylor Expansion Approximation Network for Pansharpening.

IEEE transactions on image processing : a publication of the IEEE Signal Processing Society·2026
Same journal

Noisy Tensor Completion for Sparse-Aperture Microwave Imaging in Distributed MIMO Radar Networks.

IEEE transactions on image processing : a publication of the IEEE Signal Processing Society·2026
Same journal

CGGS: Consistency-Augmented Geometric Gaussian Splatting for Ego-centric 3D Scene Generation.

IEEE transactions on image processing : a publication of the IEEE Signal Processing Society·2026

Area of Science:

  • Signal Processing
  • Machine Learning

Background:

  • Vector quantization (VQ) is crucial for data compression and pattern recognition.
  • The generalized Lloyd algorithm (GLA) is a standard method for VQ codebook generation but can be computationally intensive.
  • Efficient search for the minimum distance partition (MDP) is key to reducing GLA's computational cost.

Purpose of the Study:

  • To propose a fast search method for vector quantization.
  • To significantly reduce the computation time of codebook training using the GLA.
  • To maintain the quality of the generated codebook.

Main Methods:

  • A novel search strategy is introduced that leverages the property of vector placement in successive GLA iterations.
  • The search space for the MDP is restricted to a small subset of partitions, including the previous iteration's MDP.

Related Experiment Videos

  • Efficient schemes for creating this subset are proposed.

    • Main Results:

    • The proposed method achieves significant speedups in codebook training computation time, with improvements ranging from 6.6x to 70.4x across different codebook sizes (N=16 to 2048).
    • The generated codebooks are identical to those produced by the standard GLA.
    • Combining the method with existing techniques further enhances computational efficiency.

    Conclusions:

    • The proposed fast search method offers a computationally efficient alternative for VQ codebook training.
    • It requires only minor modifications to the GLA and modest additional memory.
    • The method demonstrates substantial improvements in training speed without compromising codebook quality.