Related Experiment Video
Updated: Jul 17, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
Lightweight and efficient dual-path fusion network for iris segmentation
Songze Lei1, Aokui Shan1, Bo Liu1
1School of Computer Science and Engineering, Xi'an Technological University, Xi'an, Shaanxi, China.
Scientific Reports
|August 28, 2023
Summary
A new lightweight iris segmentation network addresses deep learning limitations. This efficient model significantly improves accuracy and reduces computational resources, making iris segmentation more accessible.
Area of Science:
- Computer Vision
- Biometrics
- Deep Learning
Background:
- Current deep learning iris segmentation methods suffer from high parameter counts, computational demands, and large storage requirements.
- These limitations hinder the practical application of advanced iris recognition technologies.
Purpose of the Study:
- To propose a lightweight and efficient iris segmentation network.
- To overcome the resource-intensive nature of existing deep learning models for iris segmentation.
Main Methods:
- A dual-path fusion network model based on U-net architecture.
- Integration of deep semantic and shallow context information using depth-wise separable convolution.
- Introduction of a novel attention mechanism to enhance feature extraction.
Main Results:
- The proposed approach achieved average improvements of 15% in Mean Intersection over Union (MIoU) and 9% in F1 score compared to traditional methods.
- Achieved average improvements of 1.5% in MIoU and 2.5% in F1 score compared to U-net.
- Significantly reduced computation (80%), parameters (90%), and storage (99%) compared to U-net, with an average runtime of 0.02s.
Conclusions:
- The proposed lightweight network offers a superior balance between performance and efficiency for iris segmentation.
- This model presents a viable alternative for resource-constrained environments requiring accurate iris segmentation.
