Manuscript received March 22, 2025; revised April 24, 2025; accepted June 3, 2025; published October 17, 2025.

Abstract—This study analyzes traditional and deeplearning- based feature detectors and descriptors for Unmanned Aerial Vehicle (UAV) image stitching under perspective distortion. Traditional methods, including Scale- Invariant Feature Transform (SIFT), Speeded-Up Robust Features (SURF), and Accelerated-KAZE (AKAZE), achieve high keypoint detection and matching accuracy but struggle with geometric distortions, leading to artifacts such as ghosting and misalignment in complex UAV scenes. Deep-learning-based approaches, such as SuperPoint, Adaptive and Lightweight Key Point Detector (ALIKED), and Deep Image Structure Keypoints (DISK), offer superior alignment accuracy and visual coherence by detecting fewer but more robust keypoints. DISK and ALIKED demonstrate high spatial consistency, reducing perceptual artifacts as validated by Perceptual Image Quality Evaluator (PIQE), Naturalness Image Quality Evaluator (NIQE), and Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) metrics. The results indicate that deep-learning-based detectors outperform traditional methods in UAV image stitching under perspective distortions. Future work will explore hybrid models that combine the strengths of both approaches to enhance stitching accuracy and computational efficiency.

Keywords—stitching, feature detector and descriptor, perspective distortion, Unmanned Aerial Vehicle (UAV) images, no-reference metrics

Cite: Mark P. B. Pacot and Nelson Marcos, "Comprehensive Analysis of Feature Detectors and Descriptors for Stitching UAV Images with Perspective Distortion," Journal of Image and Graphics, Vol. 13, No. 5, pp. 528-539, 2025.

Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC-BY-4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.