End-to-End Exposure Fusion Using Convolutional Neural Network

Jinhua WANG; Weiqiang WANG; Guangmei XU; Hongzhe LIU

doi:10.1587/transinf.2017EDL8173

End-to-End Exposure Fusion Using Convolutional Neural Network

Jinhua WANG, Weiqiang WANG, Guangmei XU, Hongzhe LIU

Full Text Views

0

Share
Cite this

Summary :

In this paper, we describe the direct learning of an end-to-end mapping between under-/over-exposed images and well-exposed images. The mapping is represented as a deep convolutional neural network (CNN) that takes multiple-exposure images as input and outputs a high-quality image. Our CNN has a lightweight structure, yet gives state-of-the-art fusion quality. Furthermore, we know that for a given pixel, the influence of the surrounding pixels gradually increases as the distance decreases. If the only pixels considered are those in the convolution kernel neighborhood, the final result will be affected. To overcome this problem, the size of the convolution kernel is often increased. However, this also increases the complexity of the network (too many parameters) and the training time. In this paper, we present a method in which a number of sub-images of the source image are obtained using the same CNN model, providing more neighborhood information for the convolution operation. Experimental results demonstrate that the proposed method achieves better performance in terms of both objective evaluation and visual quality.

Publication: IEICE TRANSACTIONS on Information Vol.E101-D No.2 pp.560-563

Publication Date: 2018/02/01

Publicized: 2017/11/22

Online ISSN: 1745-1361

DOI: 10.1587/transinf.2017EDL8173

Type of Manuscript: LETTER

Category: Image Recognition, Computer Vision

Authors

Jinhua WANG
  University of Chinese Academy of Sciences,Beijing Union University
Weiqiang WANG
  University of Chinese Academy of Sciences
Guangmei XU
  Beijing Union University
Hongzhe LIU
  Beijing Key Laboratory of Information Service Engineering

Keyword

exposure fusion, convolutional neural networks, fusion rule, activity level measurement

Cite this

Copy

Jinhua WANG, Weiqiang WANG, Guangmei XU, Hongzhe LIU, "End-to-End Exposure Fusion Using Convolutional Neural Network" in IEICE TRANSACTIONS on Information, vol. E101-D, no. 2, pp. 560-563, February 2018, doi: 10.1587/transinf.2017EDL8173.
Abstract: In this paper, we describe the direct learning of an end-to-end mapping between under-/over-exposed images and well-exposed images. The mapping is represented as a deep convolutional neural network (CNN) that takes multiple-exposure images as input and outputs a high-quality image. Our CNN has a lightweight structure, yet gives state-of-the-art fusion quality. Furthermore, we know that for a given pixel, the influence of the surrounding pixels gradually increases as the distance decreases. If the only pixels considered are those in the convolution kernel neighborhood, the final result will be affected. To overcome this problem, the size of the convolution kernel is often increased. However, this also increases the complexity of the network (too many parameters) and the training time. In this paper, we present a method in which a number of sub-images of the source image are obtained using the same CNN model, providing more neighborhood information for the convolution operation. Experimental results demonstrate that the proposed method achieves better performance in terms of both objective evaluation and visual quality.
URL: https://globals.ieice.org/en_transactions/information/10.1587/transinf.2017EDL8173/_p

Copy

@ARTICLE{e101-d_2_560,
author={Jinhua WANG, Weiqiang WANG, Guangmei XU, Hongzhe LIU, },
journal={IEICE TRANSACTIONS on Information},
title={End-to-End Exposure Fusion Using Convolutional Neural Network},
year={2018},
volume={E101-D},
number={2},
pages={560-563},
abstract={In this paper, we describe the direct learning of an end-to-end mapping between under-/over-exposed images and well-exposed images. The mapping is represented as a deep convolutional neural network (CNN) that takes multiple-exposure images as input and outputs a high-quality image. Our CNN has a lightweight structure, yet gives state-of-the-art fusion quality. Furthermore, we know that for a given pixel, the influence of the surrounding pixels gradually increases as the distance decreases. If the only pixels considered are those in the convolution kernel neighborhood, the final result will be affected. To overcome this problem, the size of the convolution kernel is often increased. However, this also increases the complexity of the network (too many parameters) and the training time. In this paper, we present a method in which a number of sub-images of the source image are obtained using the same CNN model, providing more neighborhood information for the convolution operation. Experimental results demonstrate that the proposed method achieves better performance in terms of both objective evaluation and visual quality.},
keywords={},
doi={10.1587/transinf.2017EDL8173},
ISSN={1745-1361},
month={February},}

Copy

TY - JOUR
TI - End-to-End Exposure Fusion Using Convolutional Neural Network
T2 - IEICE TRANSACTIONS on Information
SP - 560
EP - 563
AU - Jinhua WANG
AU - Weiqiang WANG
AU - Guangmei XU
AU - Hongzhe LIU
PY - 2018
DO - 10.1587/transinf.2017EDL8173
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E101-D
IS - 2
JA - IEICE TRANSACTIONS on Information
Y1 - February 2018
AB - In this paper, we describe the direct learning of an end-to-end mapping between under-/over-exposed images and well-exposed images. The mapping is represented as a deep convolutional neural network (CNN) that takes multiple-exposure images as input and outputs a high-quality image. Our CNN has a lightweight structure, yet gives state-of-the-art fusion quality. Furthermore, we know that for a given pixel, the influence of the surrounding pixels gradually increases as the distance decreases. If the only pixels considered are those in the convolution kernel neighborhood, the final result will be affected. To overcome this problem, the size of the convolution kernel is often increased. However, this also increases the complexity of the network (too many parameters) and the training time. In this paper, we present a method in which a number of sub-images of the source image are obtained using the same CNN model, providing more neighborhood information for the convolution operation. Experimental results demonstrate that the proposed method achieves better performance in terms of both objective evaluation and visual quality.
ER -