Author ORCID Identifier
https://orcid.org/0000-0002-9633-9652
Date Available
12-31-2025
Year of Publication
2024
Document Type
Doctoral Dissertation
Degree Name
Doctor of Philosophy (PhD)
College
Arts and Sciences
Department/School/Program
Statistics
Advisor
Dr. Richard Charnigo
Abstract
This dissertation explores advanced methodologies for edge detection and image denoising through the application of both traditional non-parametric methods and modern self-supervised deep learning techniques. Beginning with non-parametric approaches, we refine surface fitting and jump detection criteria to enhance the detection of discontinuous regression surfaces in grayscale images. These foundational techniques are extended to color images, with analyses across RGB and CIELAB color spaces to improve edge detection accuracy. We then introduce a self-supervised neural network model that integrates Masked Modeling into the Bi-Directional Cascade Network (BDCN) framework. This approach shows the potential of reducing the dependency on annotated data while maintaining effective performance in edge detection tasks. By integrating self-supervised techniques, the model demonstrates adaptability across diverse visual scenarios. Masked modeling shows potential for enhancing qualitative edge detection, yielding visually promising results in certain examples, while maintaining consistency across different tasks.
Digital Object Identifier (DOI)
https://doi.org/10.13023/etd.2024.477
Funding Information
Research for Chapters 1 and 2 was sponsored by the Army Research Laboratory and was accomplished under Grant Number W911NF-17-1-0040
Recommended Citation
Xu, Jiacheng, "From Non-Parametric Methods to Self-Supervised Learning: Applications in Edge Detection and Image Denoising" (2024). Theses and Dissertations--Statistics. 81.
https://uknowledge.uky.edu/statistics_etds/81
