Related Experiment Video
Updated: Jun 21, 2026
10:16
Digital Inline Holographic Microscopy (DIHM) of Weakly-scattering Subjects
Published on: February 8, 2014
Deep homography-based image stitching with enhanced small object detection
Jicheng Cong1, Xiaotian Ran2, Chunbin Qin3
1School of Animation, Huanghuai University, Zhumadian, 463000, Henan, China.
Scientific Reports
|June 19, 2026
Summary
This study introduces a new image stitching method using YOLOv8 for better small object detection and complex backgrounds. The approach enhances feature extraction and deep homography estimation, significantly improving stitching accuracy and visual continuity.
Area of Science:
- Computer Vision
- Deep Learning
- Image Processing
Background:
- Traditional image stitching methods struggle with small objects and complex backgrounds.
- Limitations include reduced accuracy and visual artifacts in challenging image conditions.
Purpose of the Study:
- To develop a novel image stitching methodology overcoming limitations of traditional techniques.
- To enhance small object detection and improve feature alignment in complex scenes.
Main Methods:
- Leveraging YOLOv8 as a feature extractor with low-resolution feature map enhancement.
- Utilizing multi-scale feature extraction with Spatial Pyramid Pooling.
- Implementing a deep learning-based homography estimation module with a weighted loss function.
Main Results:
- Achieved higher Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) on UDIS and MVS-Synth datasets.
- Demonstrated reduced model size (188 MB) and computational complexity (14.5 GFLOPs).
- Ablation studies confirmed the effectiveness of individual modules.
Conclusions:
- The proposed YOLOv8-based image stitching method significantly improves accuracy and efficiency.
- The approach effectively handles small objects and complex backgrounds, reducing stitching artifacts.
- Future work aims to extend the model for real-time applications in dynamic environments.
