Efficient Deep Neural Network for Restoring Image Intensity

다국어 초록 (Multilingual Abstract)

In order to maximize the image quality when using existing image contents in the latest display devices, it is necessary to improve the resolution and intensity of the images. In this paper, we propose an efficient deep neural network to restore the image intensity when there are too few bits per pixel to provide more intensity. We investigate an efficient implementation and training method for U-net to maximize the performance of restoring image intensity. We show that we can significantly improve the perceptual quality of the restored image by using VGG loss as well as MSE loss to train U-net. The perceptual loss of images can be efficiently dealt with by using VGG loss. The convergence of the proposed method is analyzed, and extensive computer simulations show that the proposed method significantly improves the perceptual quality of the restored image.

목차 (Table of Contents)

• Abstract
• 1. Introduction
• 2. Restoration of Image Intensity and Deep Learning
• 3. Proposed Method for Restoring Image Intensity Based on Deep Learning
• 4. Performance Evaluation
• 5. Conclusion
• References

참고문헌 (Reference)

1 "http://cocodataset.org"

2 K. Simonyan, "Very deep convolutional networks for large-scale image recognition" 2015

3 O. Ronneberger, "U-Net:Convolutional networks for biomedical image segmentation" 234-241, 2015

4 H. Dong, "TensorLayer: A versatile library for efficient deep learning development" 1201-1204, 2017

5 K. S. Heo, "Restoration of image intensity using deep learning" 2019

6 C. Ledig, "Photo-realistic single image super-resolution using a generative adversarial network" 105-114, 2017

7 J. Johnson, "Perceptual losses for real-time style transfer and superresolution" 694-711, 2016

8 W. Shi, "Is the deconvolution layer the same as a convolutional layer?"

9 C. Dong, "Image super-resolution using deep convolutional networks" 38 (38): 295-307, 2016

10 X. Hou, "Image companding and inverse halftoning using deep convolutional neural networks"

11 R. P. Kovaleski, "High-quality reverse tone mapping for a wide range of exposures" 2014

12 G. Eilertsen, "HDR image reconstruction from a single exposure using deep CNNs" 36 (36): 2017

13 G. Huang, "Densely connected convolutional networks" 4700-4708, 2017

14 K. He, "Deep residual learning for image recognition" 770-778, 2016

15 D. Kingma, "Adam: A method for stochastic optimization" 2015