Mix: A Potential Image Augmentation Method on Retinal Vessel Segmentation

Convolutional neural networks(CNN) is a powerful method to complete medical image segmentation task. It is standard practice to enhance raw inputs before feeding into CNN to avoid overfitting. In this paper, we proposed a new method for image augmentation called mix. It combines two patches of image...

Full description

Saved in:
Bibliographic Details
Published in2021 6th International Conference on Image, Vision and Computing (ICIVC) pp. 140 - 143
Main Authors Zeng, HongWei, Yi, XingWen, Liang, ShanShan
Format Conference Proceeding
LanguageEnglish
Published IEEE 23.07.2021
Subjects
Computed tomography
Convolutional neural networks
Costs
Image Augmentation
Image segmentation
Magnetic resonance imaging
Mix
Retinal Vessel Segmentation
Retinal vessels
Robustness
Online AccessGet full text

Cover

More Information
Summary:Convolutional neural networks(CNN) is a powerful method to complete medical image segmentation task. It is standard practice to enhance raw inputs before feeding into CNN to avoid overfitting. In this paper, we proposed a new method for image augmentation called mix. It combines two patches of image by averaging their pixel value, and combines the corresponding labels by logic OR operation. Our experiments achieve the area of curve(AUC) of 0.9836 on DRIVE and 0.9913 on STARE, outperforming the traditional augmentation method, which indicates that mix improves the robustness of network and can be a considerable data augmentation method on segmentation task.
DOI:10.1109/ICIVC52351.2021.9527011