Related Experiment Video
Updated: Jun 17, 2026
05:33
Three-Dimensional Shape Modeling and Analysis of Brain Structures
Published on: November 14, 2019
Improving shape retrieval by spectral matching and meta similarity
Amir Egozi1, Yosi Keller, Hugo Guterman
1Department of Electrical Engineering, Ben-Gurion University, Beer-Sheva, Israel. agozi@ee.bgu.ac.il
Summary
We developed new computational methods for shape retrieval, achieving 92.5% accuracy on the MPEG-7 dataset. Our techniques improve shape matching despite noise and deformation, enhancing retrieval performance.
Area of Science:
- Computer Vision
- Pattern Recognition
- Computational Geometry
Background:
- Accurate retrieval of planar shapes is crucial for various applications.
- Existing shape retrieval methods struggle with noise, articulations, and non-rigid deformations.
Purpose of the Study:
- To propose novel computational approaches for enhancing planar shape retrieval.
- To develop a robust similarity measure resilient to common shape distortions.
Main Methods:
- A geometrically motivated quadratic similarity measure optimized via spectral relaxation.
- Utilization of advanced shape descriptors and a pairwise serialization constraint.
- Introduction of a shape meta-similarity measure for improved retrieval accuracy.
Main Results:
- The proposed formulation demonstrates resilience to boundary noise, articulations, and non-rigid deformations.
- Achieved a 92.5% retrieval rate on the MPEG-7 shape dataset.
- The meta-similarity measure significantly improved overall retrieval accuracy.
Conclusions:
- The developed computational approaches offer a robust and accurate solution for planar shape retrieval.
- The geometric matching scheme and meta-similarity measure represent significant advancements in the field.
- These methods have practical implications for content-based image retrieval and computer vision tasks.
