Low light combining multiscale deep learning networks and image enhancement algorithm
Aiming at the lack of reference images for low-light enhancement tasks and the problems of color distortion, texture loss, blurred details, and difficulty in obtaining ground-truth images in existing algorithms, this paper proposes a multi-scale weighted feature low-light based on Retinex theory an...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Siberian Scientific Centre DNIT
2022-11-01
|
| Series: | Современные инновации, системы и технологии |
| Subjects: | |
| Online Access: | https://oajmist.com/index.php/12/article/view/195 |
| _version_ | 1811303726108901376 |
|---|---|
| author | Ся Ю Лин Бо Чен Синь |
| author_facet | Ся Ю Лин Бо Чен Синь |
| author_sort | Ся Ю |
| collection | DOAJ |
| description |
Aiming at the lack of reference images for low-light enhancement tasks and the problems of color distortion, texture loss, blurred details, and difficulty in obtaining ground-truth images in existing algorithms, this paper proposes a multi-scale weighted feature low-light based on Retinex theory and attention mechanism. An image enhancement algorithm is proposed. The algorithm performs multi-scale feature extraction on low-light images through the feature extraction module based on the Unet architecture, generates a high-dimensional multi-scale feature map, and establishes an attention mechanism module to highlight the feature information of different scales that are beneficial to the enhanced image, and obtain a weighted image. High-dimensional feature map, the final reflection estimation module uses Retinex theory to build a network model, and generates the final enhanced image through the high-dimensional feature map. An end-to-end network architecture is designed and a set of self-regular loss functions are used to constrain the network model, which gets rid of the constraints of reference images and realizes unsupervised learning. The final experimental results show that the algorithm in this paper maintains high image details and textures while enhancing the contrast and clarity of the image, has good visual effects, can effectively enhance low-light images, and greatly improves the visual quality. Compared with other enhanced algorithms, the objective indicators PSNR and SSIM have been improved.
|
| first_indexed | 2024-04-13T07:53:08Z |
| format | Article |
| id | doaj.art-1704d15c0b8449bca8c93ddceced2614 |
| institution | Directory Open Access Journal |
| issn | 2782-2826 2782-2818 |
| language | English |
| last_indexed | 2024-04-13T07:53:08Z |
| publishDate | 2022-11-01 |
| publisher | Siberian Scientific Centre DNIT |
| record_format | Article |
| series | Современные инновации, системы и технологии |
| spelling | doaj.art-1704d15c0b8449bca8c93ddceced26142022-12-22T02:55:28ZengSiberian Scientific Centre DNITСовременные инновации, системы и технологии2782-28262782-28182022-11-012410.47813/2782-2818-2022-2-4-0215-0232Low light combining multiscale deep learning networks and image enhancement algorithmСя Ю 0Лин Бо1Чен Синь2Sofia University, BulgariaSouth China Normal University, China Shenzhen University, China Aiming at the lack of reference images for low-light enhancement tasks and the problems of color distortion, texture loss, blurred details, and difficulty in obtaining ground-truth images in existing algorithms, this paper proposes a multi-scale weighted feature low-light based on Retinex theory and attention mechanism. An image enhancement algorithm is proposed. The algorithm performs multi-scale feature extraction on low-light images through the feature extraction module based on the Unet architecture, generates a high-dimensional multi-scale feature map, and establishes an attention mechanism module to highlight the feature information of different scales that are beneficial to the enhanced image, and obtain a weighted image. High-dimensional feature map, the final reflection estimation module uses Retinex theory to build a network model, and generates the final enhanced image through the high-dimensional feature map. An end-to-end network architecture is designed and a set of self-regular loss functions are used to constrain the network model, which gets rid of the constraints of reference images and realizes unsupervised learning. The final experimental results show that the algorithm in this paper maintains high image details and textures while enhancing the contrast and clarity of the image, has good visual effects, can effectively enhance low-light images, and greatly improves the visual quality. Compared with other enhanced algorithms, the objective indicators PSNR and SSIM have been improved. https://oajmist.com/index.php/12/article/view/195deep learning, Retinex theory, improve low light picture, image inpainting |
| spellingShingle | Ся Ю Лин Бо Чен Синь Low light combining multiscale deep learning networks and image enhancement algorithm Современные инновации, системы и технологии deep learning, Retinex theory, improve low light picture, image inpainting |
| title | Low light combining multiscale deep learning networks and image enhancement algorithm |
| title_full | Low light combining multiscale deep learning networks and image enhancement algorithm |
| title_fullStr | Low light combining multiscale deep learning networks and image enhancement algorithm |
| title_full_unstemmed | Low light combining multiscale deep learning networks and image enhancement algorithm |
| title_short | Low light combining multiscale deep learning networks and image enhancement algorithm |
| title_sort | low light combining multiscale deep learning networks and image enhancement algorithm |
| topic | deep learning, Retinex theory, improve low light picture, image inpainting |
| url | https://oajmist.com/index.php/12/article/view/195 |
| work_keys_str_mv | AT sâû lowlightcombiningmultiscaledeeplearningnetworksandimageenhancementalgorithm AT linbo lowlightcombiningmultiscaledeeplearningnetworksandimageenhancementalgorithm AT čensinʹ lowlightcombiningmultiscaledeeplearningnetworksandimageenhancementalgorithm |