DEEP UNFOLDING FOR ITERATIVE STRIPE NOISE REMOVAL

Zeshan Fayyaz, Daniel Platnick, Hannan Fayyaz, Nariman Farsad

The non-uniform photoelectric response of infrared imaging results in fixed-pattern stripe noise being superimposed on infrared images, which severely reduces image quality. Existing destriping methods struggle to concurrently remove all stripe noise artifacts, preserve image details and structures, and balance real-time performance. In this paper, we propose an innovative destriping method which takes advantage of spatial feature estimation. Our model iteratively uses neighbouring column signal correlation to remove independent column stripe noise. The proposed method allows for a more precise estimation of stripe noise to preserve scene details more accurately. Extensive experimental results demonstrate that the proposed model outperforms existing destriping methods on artificially corrupted images on both quantitative and qualitative assessments.

INSTALLATION

Python 3.8.10
Tensorflow 2.5.0
Keras 2.4.3 

DIRECTORY ORGANIZATION

1. Please download the Linnaeus 5 and BSDS images from ./Datasets/Train/BSDS500/ and ./Datasets/Train/dataset. Combine all training images and place into 'CombinedTrain' directory.

2. Please download the testing folders found in ./Datasets/Test/

Folder architecture should be as follows:

./Datasets/
./Datasets/Train/
./Datasets/Train/CombinedTrain/
./Datasets/Test/
./Datasets/Test/BSDS100/
./Datasets/Test/INFRARED100/
./Datasets/Test/Linnaeus5/
./Datasets/Test/Set12/
./Datasets/Test/Urban100/

EVALUATE MODEL ON PRETRAINED WEIGHTS

Download our pretrained model found from ./pretrained_model/destriping_model/

Download evaluate.py

Edit Line 37 to reflect the directory which contains destriping_model.h5

Edit Line 23, 26, 29, 32, and 35 to reflect the directories which contains the 5 test datasets [BSDS100, INFRARED100, Set12, Linnaeus5, Urban100]

Uncomment the TESTDIR you wish to evaluate on. Script returns degraded and predicted PSNR and SSIM values.

TRAINING THE MODEL

Download main.py

Create two directories. ./15GRU/trained_model_weights/ and ./15GRU/Results/

Edit Line 22 and 23 to reflect training dataset and test dataset.

Edit Line 24 to the directory [./15GRU/Results/] where we wish to save: 10 pdf images containing degraded and predicted testing images, 10 corresponding .txt files containing PSNR and SSIM metrics, TrainingLoss.pdf, TrainingValidationLoss.pdf, and ValidationLoss.pdf

Edit Line 229 and 279 to ./15GRU/trained_model_weights/ where we save model_checkpoint.h5 and stripe_noise_model.h5

An example of a saved .pdf image containing predicted images is as follows:

ABLATION STUDY