Exploration of Deep Methods for Cross-Domain Aerial Image Matching

• Published in: International Workshops on Image Processing Theory, Tools, and Applications pp. 1 - 5
• Main Authors: Shahid, Md, Vk, Prajwal, Haq, Inzamam-ul, Faisal, Md
• Format: Conference Proceeding
• Language: English
• Published: IEEE 16.10.2023
• Subjects: Aerial UAV; cross-platform matching; Deep learning; Dimensionality reduction; Image matching; Imaging; Performance evaluation; Propulsion; satellite image; Training
• ISSN: 2154-512X
• DOI: 10.1109/IPTA59101.2023.10320028

Image matching is an attractive area for researchers. This field has been propelled recently due to the advancement of imaging devices with multi-spectral capabilities, compute power, and the evolution of the deep learning era. Cross-domain or cross-platform matching remains challenging due to the multiple type modalities associated and the need for relevant datasets. Learning or training in one domain (e.g., spectral band) and matching across domains are the real challenges for the imaging community today. The wide availability of Google Earth imagery, the development of cross-domain matching algorithms, and high-speed devices encourage addressing this issue. This paper explored deep learning methods for a cross-domain image-matching perspective. We have exploited latent and feature space, particularly for standard deep methods with several preprocessing steps, optimizer settings, and post-processing steps. We evaluated performance over standard cross-platform dataset using metrics available in the literature.