GGADN: Guided generative adversarial dehazing network
Summary (2 min read)
1 Introduction
- Especially for haze, floating particles in haze lead to the fading and blurring of pictures, and the reduction of contrast and softness.
- At present, image dehazing research is mainly divided into two types, feature-based method and learning-based method.
- The guided module verifies the network of each layer of the generator.
3 The proposed method
- This paper presents a Guided Generative Adversarial Dehazing Network.
- The loss function is modified by using the pre-trained VGG feature and L1-regularization gradient.
- In the proposed algorithm, an end-to-end dehazing network is used to train the network to avoid image distortion or artifact which caused by the estimation of transmittance and atmospheric light value.
- They are defined from left to right as follows: – Conv1: In-channels are three.
- The GGan training method is to directly use the Adversarial loss function is expressed as: LA = 1 N N∑ i=1 log [1−D(Ii, J̃i)] (3) Where D is the discriminant network, is the output of generator G. Discriminators are input in Minibatch mode.
4 Experimental results and analysis
- To evaluate the performance of proposed model, the authors compare it with several advanced single image dehazing algorithms on synthetic datasets and real scene images.
- Among them, 3000 synthetic images are randomly selected and 300 test images are extracted.
- 2 Implementation Keras deep learning architecture is used to train the model, RMSprop algorithm is used to optimize the model parameters, and the training epochs is 100.
- Compared with other dehazing methods (DCP, DehazeNet, MSCNN and AODNet ), the advantages of proposed method are that the detail information of dehazing image is preserved completely, the color recovery is more natural, and the degree of dehazing is moderate.
5 Conclusions
- This paper presents a Guided Generative Adversarial Dehazing Network.
- In GGADN the generator and discriminator architecture are modified.
- The synthetic data set is trained by end-to-end training neural network.
- Sigmoid function is introduced to the last layer of discriminator for feature mapping.
- The first one is to explore the clear image without reference for training the dehazing network [45,46,51,52].
Did you find this useful? Give us your feedback
Citations
References
3,668 citations
2,293 citations
1,880 citations
1,185 citations
Related Papers (5)
Frequently Asked Questions (17)
Q2. What is the training of the network?
Network training is based on the pre-trained VGG feature model and L1-regularized gradient prior which is developed by new loss function parameters.
Q3. What is the purpose of the proposed algorithm?
In the proposed algorithm, an end-to-end dehazing network is used to train the network to avoid image distortion or artifact which caused by the estimation of transmittance and atmospheric light value.
Q4. How many channels are used in the GGan training method?
The GGan training method is to directly use the Adversarial loss function is expressed as:LA = 1NN∑i=1log [1−D(Ii, J̃i)] (3)Where D is the discriminant network, is the output of generator G. Discriminators are input in Minibatch mode.
Q5. What is the purpose of the generator?
The generator introduces the skip connection of symmetry layer in ”RESNET” and ”u-net” models to break through the bottleneck of information redundancy in the process of dehazing.
Q6. How is the cycle haze agent enhanced?
In order to improve the quality of texture information recovery and produce a better visual haze free image, the CycleGan agent is enhanced by combining cycle consistency and perceptual loss.
Q7. What is the main reason for the haze?
Especially for haze, floating particles in haze lead to the fading and blurring of pictures, and the reduction of contrast and softness.
Q8. What is the purpose of the paper?
In order to evaluate the algorithm objectively, this paper uses Peak Signal to Noise Ratio (PSNR) [43] and Structural Similarity index (SSIM) [44] as objective evaluation indexes.
Q9. What is the value of dark channel in haze free image?
He found that the value of dark channel in haze free image is close to zero, and then it can be used to estimate the transmission image.
Q10. What is the transmittance of the atmosphere?
When the composition of the atmosphere is uniform, that is, A(x) is constant, the transmittance can be expressed as:t(x) = e−βd(x)
Q11. What is the main problem with the Gan network?
the network structure of Gan is unstable in the training process, and some artifacts such as noise and color shift are often produced in the synthetic image.
Q12. What is the purpose of image defogging?
2.1 Atmospheric Scattering ModelThe purpose of image defogging is to restore a clear image from the blurred image corroded by haze or smoke.
Q13. What are the advantages of proposed method?
Compared with other dehazing methods (DCP, DehazeNet, MSCNN and AODNet ), the advantages of proposed method are that the detail information of dehazing image is preserved completely, the color recovery is more natural, and the degree of dehazing is moderate.
Q14. What are the advantages of the proposed method?
AODNet method can generate a clear image as shown in the fifth image, but compared with the dehazing image generated by proposed algorithm, there are more details of unclear objects, the color of dehazing image is dark, and the sky part of the image has slight color distortion.
Q15. How many training sets are used to train the model?
In image processing, the whole large training set is divided into several small training sets to improve the computational efficiency and help to train the model quickly.
Q16. Why is there still a gap between the real scene and the synthetic haze image?
Because there is still a gap between the real scene and the synthetic haze image in visual perception, in order to verify the generalization ability of proposed method, the proposed method is compared with DCP, DehazeNet, MSCNN and AODNet in the natural scene images.
Q17. How can the authors predict the transmission image?
Ren [22] proposed a multi-scale convolutional neural network mscnn, which can accurately predict the transmission image through two different scale network models.