Related Experiment Video
Updated: Jul 4, 2026
08:27
Image Recognition and Parameter Analysis of Concrete Vibration State Based on Support Vector Machine
Published on: January 5, 2024
Inference of segmented color and texture description by tensor voting.
1Department of Computer Science, Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong. leojia@cs.ust.hk
Summary
This study introduces a novel tensor voting method to automatically restore missing color and texture in damaged 2D images. The technique effectively infers data for 2D and 3D applications, even with noise and occlusion.
Area of Science:
- Computer Vision
- Image Processing
- Data Synthesis
Background:
- Damaged 2D images often suffer from missing color and texture information.
- Existing methods struggle with occlusion, noise, and incomplete data.
Purpose of the Study:
- To develop a robust method for automatically inferring missing color and texture in damaged 2D images.
- To generalize this approach for range and 3D data with occlusions and noise.
Main Methods:
- Utilizes (N)D tensor voting (N > 3) to translate texture into adaptive tensors.
- Employs a two-step process: segmentation extrapolation and non-iterative color synthesis within segments.
- Incorporates tensor scale analysis for automatic adaptation to different feature scales.
Main Results:
- Successfully infers missing color and texture information from damaged 2D images.
- Demonstrates effectiveness in generalizing to range and 3D data with occlusions and noise.
- Achieves complete segmentation and synthesizes missing colors effectively within identified segments.
Conclusions:
- The proposed (N)D tensor voting method offers a robust solution for image data completion.
- The approach is versatile, handling various data types and imperfections effectively.
- This technique significantly enhances the quality and completeness of damaged image data.
