zhiyong-zhou / irissegment

Iris segmentation using feature channel optimization for noisy environments

Iris-segmentation

Iris segmentation using feature channel optimization for noisy environments

Requirements

Tensorflow 1.4.0
Keras 2.2.0
Python 3.5

Results

R stands for recall rate, P stands for precision, and F-measure is a combination of the two. (unit: %)

Dataset	F	R	P	Error rate(%)
CASIA	98.11	97.96	98.27	0.81
IITD	97.84	97.78	97.91	0.98

Data

We use CASIA V4.0 Interval (Abbr. CASIA) dataset, and the IIT Delhi v1.0 (Abbr. IITD) dataset. We provide a noisy dataset with Gaussian noise. When training the model, we use a '.npy' file of the dataset.

The weights we provide are the training results for Gaussian noise. In 'Model/CAV/gaussianNoise/model_new.hdf5' or 'Model/IITD/gaussianNoise/model_new.hdf5'

Run on GPU

• To test the model, you can run

python test_predict.py

In 'Model/CAV', you can see the segmentation results.

• In order to measure the performance of the model with the RPF metric, you can run

python error_RPF.py

• To train the model, you can run

python model.py

The training results will be written to Model/CAV

Citation

Please cite this paper if you think it is useful for you.
Title: Iris segmentation using feature channel optimization for noisy environments
Author: Kangli Hao · Guorui Feng · Yanli Ren · Xinpeng Zhang