"Frameworks for multi-source image search using latent space feature ma" by Praveen Vijayakumar Phatate

Date of Award

2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

Committee Chair

Tathagata Mukherjee

Committee Member

Sundar Christopher

Committee Member

Huaming Zhang

Committee Member

Vineetha Menon

Committee Member

Chaity Banerjee

Research Advisor

Tathagata Mukherjee

Subject(s)

Image processing--Digital techniques, Pattern recognition systems, Computer vision, Artificial intelligence

Abstract

In this dissertation we study frameworks for image matching for three distinct application scenarios: the first relates to positioning in GPS agnostic environments, the second relates to matching and retrieving similar images in an indoor environment sourced from multiple uncalibrated cameras and the third relates to matching and retrieving satellite images across multiple uncalibrated satellite cameras. We propose three different frameworks the first one based on the idea of transformers, the second one based on the idea of autoencoders and the third one based on the idea of Siamese networks. We report extensive experiments with novel datasets for each of these scenarios. For the first we use a novel high altitude aerial image dataset obtained from Leon county GIS in Florida and a publicly available dataset curated from Google Earth, for the second we curate our own dataset at UAH and use it for our experiments. Finally, for the third we use dataset from Maxar and Planet obtained at the courtesy of NASA.

Available for download on Tuesday, May 05, 2026

Share

COinS