研究目的
To develop an efficient and robust iris segmentation algorithm using deep learning for accurate pupillary and limbus boundary localization in iris recognition systems.
研究成果
The proposed iris segmentation algorithm achieves high accuracy (95.49%) and efficiency (0.06 seconds per eye) on a challenging database, outperforming previous methods. It is suitable for real-time applications and mobile devices due to its small model size and fast processing. Future work includes further speed improvements and implementation on mobile platforms.
研究不足
The algorithm may be sensitive to extreme lighting conditions or severe obstructions not covered in the database. Computational requirements might still be high for very low-end mobile devices, and the model's performance could degrade with images from different sensors or environments.
1：Experimental Design and Method Selection:
The study combines learning-based (Faster R-CNN and Gaussian mixture model) and edge-based algorithms for iris segmentation. It involves designing a shallow Faster R-CNN for eye detection, using GMM for pupillary region fitting, and employing MIGREP and boundary point selection for limbus boundary estimation.
2：Sample Selection and Data Sources:
The CASIA-Iris-Thousand database with 20,000 iris images from 1,000 subjects is used for training and testing. Images include obstructions like glasses and specular reflections.
3：List of Experimental Equipment and Materials:
A personal computer with 3.4-GHz CPUs and GTX 1080 GPU, MATLAB R2018a software, and the IKEMB-100 camera for image capture.
4：4-GHz CPUs and GTX 1080 GPU, MATLAB R2018a software, and the IKEMB-100 camera for image capture.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Steps include eye detection with Faster R-CNN, pupillary region fitting with GMM, boundary point selection, and circle fitting for boundaries. Performance is evaluated using a radial difference integration method.
5：Data Analysis Methods:
Accuracy is measured using precision, recall, and a new evaluation method based on radial difference. Statistical analysis includes comparison with previous methods.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
