CNN-Based Restoration of a Single Face Image Degraded by Atmospheric Turbulence

Yasarla, Rajeev; Patel, Vishal M.

doi:10.1109/tbiom.2022.3169697

Cited by 18 publications

(4 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The performance of the model stagnates or even decreases when the number of layers of the CNN model increases. This is the problem of degradation [ 22 ].…”

Section: Methods’ Resultsmentioning

confidence: 99%

ROENet: A ResNet-Based Output Ensemble for Malaria Parasite Classification

2022

View full text Add to dashboard Cite

(1) Background: People may be infected with an insect-borne disease (malaria) through the blood input of malaria-infected people or the bite of Anopheles mosquitoes. Doctors need a lot of time and energy to diagnose malaria, and sometimes the results are not ideal. Many researchers use CNN to classify malaria images. However, we believe that the classification performance of malaria parasites can be improved. (2) Methods: In this paper, we propose a novel method (ROENet) to automatically classify malaria parasite on the blood smear. The backbone of ROENet is the pre-trained ResNet-18. We use randomized neural networks (RNNs) as the classifier in our proposed model. Three RNNs are used in ROENet, which are random vector functional link (RVFL), Schmidt neural network (SNN), and extreme learning machine (ELM). To improve the performance of ROENet, the results of ROENet are the ensemble outputs from three RNNs. (3) Results: We evaluate the proposed ROENet by five-fold cross-validation. The specificity, F1 score, sensitivity, and accuracy are 96.68 ± 3.81%, 95.69 ± 2.65%, 94.79 ± 3.71%, and 95.73 ± 2.63%, respectively. (4) Conclusions: The proposed ROENet is compared with other state-of-the-art methods and provides the best results of these methods.

show abstract

“…The performance of the model stagnates or even decreases when the number of layers of the CNN model increases. This is the problem of degradation [ 22 ].…”

Section: Methods’ Resultsmentioning

confidence: 99%

ROENet: A ResNet-Based Output Ensemble for Malaria Parasite Classification

2022

View full text Add to dashboard Cite

show abstract

“…In [56], the effects of atmospheric turbulence on face recognition are studied, where atmospheric distortions are found to significantly affect face recognition performance. Other works have developed upstream image restoration for atmospheric turbulence [39], [75], [76]. Image restoration methods focus on image-based metrics such as PNSR, not recognition.…”

Section: Related Workmentioning

confidence: 99%

“…Furthermore, IJB-S is not a strictly long-range dataset. To overcome the lack of available long-range data, some prior works have used simulated atmospheric turbulence as a proxy for real data [56], [75], [76]. However, the effectiveness of simulated atmospherics for face recognition has not been validated because, as mentioned before, there is no real data for validation.…”

Section: A Long Distance Recapture Datamentioning

confidence: 99%

DaliID: Distortion-Adaptive Learned Invariance for Identification—A Robust Technique for Face Recognition and Person Re-Identification

Robbins,

Bertocco,

Boult

2024

IEEE Access

View full text Add to dashboard Cite

In real-world applications, face recognition and person re-identification are subject to image degradations such as motion blur, atmospheric turbulence, or upsampling artifacts-which are known to lower performance. This work directly addresses challenges in low-quality scenarios with 1) practical, novel updates to training and inference, which improve robustness to realistic distortions in face recognition and person re-identification and 2) new datasets for long-distance recognition. We propose a method that progressively learns from images prone to soft to strong distortions caused mainly by atmospheric turbulence. The method has a novel distortion loss to improve robustness, which is empirically shown to be highly effective in low-quality scenarios. Two further strategies are proposed to integrate distortion augmentation while also retaining the highest performance in high-quality scenarios. First, during training, an adaptive weighting schedule, which leverages the construction of different levels of distortion augmentation, is used to train the model in an easy-to-hard manner. The second, at inference, is a magnitude-weighted fusion of features from the parallel models used to retain the highest robustness across both high-quality and low-quality imagery. Different from prior art, our model does not leverage any image restoration or style transfer technique, and we are the first to employ explicit distortion weighting during training and evaluation. Our model achieves the best performances compared to prior works on face recognition and person re-identification benchmarks, including IJB-S, TinyFace, DeepChange, MSMT17, and our novel long-distance datasets.

show abstract

“…Subsequent work of Lou et al [20] followed the blur-then-tilt model, and this tradition continues until today where many of the latest deep learning based approaches also use the blurthen-tilt model. For example, Lau et al [21], Nair and Patel [22], Yasarla and Patel [23], and Lau and Lui [24], [25] all explicitly or implicitly use the blur-then-tilt model.…”

Section: Introductionmentioning

confidence: 99%

Tilt-then-Blur or Blur-then-Tilt? Clarifying the Atmospheric Turbulence Model

Chan¹

2022

Preprint

View full text Add to dashboard Cite

Imaging at a long distance often requires advanced image restoration algorithms to compensate for the distortions caused by atmospheric turbulence. However, unlike many standard restoration problems such as deconvolution, the forward image formation model of the atmospheric turbulence does not have a simple expression. Thanks to the Zernike representation of the phase, one can show that the forward model is a combination of tilt (pixel shifting due to the linear phase terms) and blur (image smoothing due to the high order aberrations).Confusions then arise between the ordering of the two operators. Should the model be tilt-then-blur, or blur-then-tilt? Some papers in the literature say that the model is tilt-thenblur, whereas more papers say that it is blur-then-tilt. This paper clarifies the differences between the two and discusses why the tilt-then-blur is the correct model. Recommendations are given to the research community.

show abstract

CNN-Based Restoration of a Single Face Image Degraded by Atmospheric Turbulence

Cited by 18 publications

References 51 publications

ROENet: A ResNet-Based Output Ensemble for Malaria Parasite Classification

ROENet: A ResNet-Based Output Ensemble for Malaria Parasite Classification

DaliID: Distortion-Adaptive Learned Invariance for Identification—A Robust Technique for Face Recognition and Person Re-Identification

Tilt-then-Blur or Blur-then-Tilt? Clarifying the Atmospheric Turbulence Model

Contact Info

Product

Resources

About