Towards More Accurate and Complete Heterogeneous Iris Segmentation Using a Hybrid Deep Learning Approach

Area of Science:

• Computer Vision
• Medical Imaging
• Deep Learning

Background:

• Accurate iris segmentation is vital for computer-aided ophthalmic disease diagnosis.
• Varying image quality from different sensors presents challenges for segmenting heterogeneous iris databases.
• Current Convolutional Neural Network (CNN) based methods struggle with global semantic information for precise iris segmentation.

Purpose of the Study:

• To develop an advanced deep learning approach for robust iris segmentation in heterogeneous image datasets.
• To overcome the limitations of CNNs in capturing long-range dependencies for improved iris segmentation accuracy.

Main Methods:

• A hybrid deep learning model combining Vision Transformer (Swin Transformer) and CNNs was proposed.
• A multiscale feature information extraction module (MFIEM) was developed for granular spatial information capture.
• A channel attention mechanism module (CAMM) was integrated to enhance iris region discriminability.

Main Results:

• The proposed hybrid network demonstrated superior performance on a multisource heterogeneous iris database.
• Experimental results showed a significant advantage over existing state-of-the-art iris segmentation methods.
• The integration of Swin Transformer and CNNs effectively addressed limitations in capturing global semantic information.

Conclusions:

• The novel hybrid deep learning approach offers a significant advancement in iris segmentation technology.
• This method provides a more accurate and reliable tool for preprocessing in ophthalmic disease diagnosis.
• The findings highlight the potential of combining transformers and CNNs for complex image segmentation tasks.