Estimation of ambient light and transmission map with common convolutional architecture

Shin, Young-Sik; Cho, Younggun; Pandey, Gaurav; Kim, Ayoung

doi:10.1109/oceans.2016.7761342

Cited by 63 publications

(63 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Deep learning techniques in the context of underwater robots have been mainly used with an opposite goal in mind: underwater color correction. In [37] a CNN (Convolutional Neural Network) is trained to estimate the ambient light and thus to dehaze the image. A more similar approach to style transfer is faced in [38] where the cross-domain relations between air and underwater images are learnt.…”

Section: Style Transfermentioning

confidence: 99%

See 1 more Smart Citation

Generation and Processing of Simulated Underwater Images for Infrastructure Visual Inspection with UUVs

Álvarez-Tuñón

Jardón

Balaguer

2019

Sensors

View full text Add to dashboard Cite

The development of computer vision algorithms for navigation or object detection is one of the key issues of underwater robotics. However, extracting features from underwater images is challenging due to the presence of lighting defects, which need to be counteracted. This requires good environmental knowledge, either as a dataset or as a physic model. The lack of available data, and the high variability of the conditions, makes difficult the development of robust enhancement algorithms. A framework for the development of underwater computer vision algorithms is presented, consisting of a method for underwater imaging simulation, and an image enhancement algorithm, both integrated in the open-source robotics simulator UUV Simulator. The imaging simulation is based on a novel combination of the scattering model and style transfer techniques. The use of style transfer allows a realistic simulation of different environments without any prior knowledge of them. Moreover, an enhancement algorithm that successfully performs a correction of the imaging defects in any given scenario for either the real or synthetic images has been developed. The proposed approach showcases then a novel framework for the development of underwater computer vision algorithms for SLAM, navigation, or object detection in UUVs.

show abstract

Section: Style Transfermentioning

confidence: 99%

“…RGHS (Relative Global Histogram Stretching) [45] and RD (Rayleigh Distribution) [46] perform both contrast and color correction. The recent advances in deep learning have also been applied for both restoration and enhancement techniques [37,38].…”

Section: Underwater Image Enhancementmentioning

confidence: 99%

Generation and Processing of Simulated Underwater Images for Infrastructure Visual Inspection with UUVs

Álvarez-Tuñón

Jardón

Balaguer

2019

Sensors

View full text Add to dashboard Cite

show abstract

“…On the other hand, underwater image restoration methods aim to recover clear images by exploiting the optical imaging model. The most important task for restoration is to estimate two key model parameters, i.e., transmission and ambient light, which are usually estimated either by prior-based approaches [6,[18][19][20][21][22][23][24][25] or by learning-based approaches [26][27][28][29][30]. The prior-based approaches heavily depend on the reliability of certain prior information, such as dark channel prior [18][19][20][21], red channel prior [6], haze-line prior [25] and so on.…”

Section: Introductionmentioning

confidence: 99%

“…Thus, a mismatch between the adopted prior and the target scene may incur significant estimation error, and consequently recover distorted results [23]. By contrast, the learning-based approaches aim to obtain more robust and accurate estimation by exploring the relations between the underwater images and the corresponding parameters in a data-driven manner, such as [26][27][28][29][30]. To this end, it is essential to have a suitable training dataset and an efficient neural network that can be trained to learn such relations.…”

Section: Introductionmentioning

confidence: 99%

Underwater Image Restoration Based on a Parallel Convolutional Neural Network

Wang

et al. 2019

Remote Sensing

View full text Add to dashboard Cite

Restoring degraded underwater images is a challenging ill-posed problem. The existing prior-based approaches have limited performance in many situations due to the reliance on handcrafted features. In this paper, we propose an effective convolutional neural network (CNN) for underwater image restoration. The proposed network consists of two paralleled branches: a transmission estimation network (T-network) and a global ambient light estimation network (A-network); in particular, the T-network employs cross-layer connection and multi-scale estimation to prevent halo artifacts and to preserve edge features. The estimates produced by these two branches are leveraged to restore the clear image according to the underwater optical imaging model. Moreover, we develop a new underwater image synthesizing method for building the training datasets, which can simulate images captured in various underwater environments. Experimental results based on synthetic and real images demonstrate that our restored underwater images exhibit more natural color correction and better visibility improvement against several state-of-the-art methods.

show abstract

“…The priors are directly computed or trained earlier; however, the assumption of the priors easily breaks on a real robot operating situations, such as smoke regions and underwater. Shin, Cho, Pandey, and Kim () proposed a CNN‐based method for the estimation of a colored ambient light and transmission map. Also, Li, Skinner, Eustice, and Johnson‐Roberson () presented waterGAN, which is used for generating haze images.…”

Section: Introductionmentioning

confidence: 99%

Channel invariant online visibility enhancement for visual SLAM in a turbid environment

Cho

Kim

2018

Journal of Field Robotics

Self Cite

View full text Add to dashboard Cite

This paper presents a real‐time and channel‐invariant visibility enhancement algorithm using a hybrid image enhancement approach. The proposed method is initially motivated by an underwater visual simultaneous localization and mapping (SLAM) failure in a turbid medium. The environments studied contain various particles and are dominated by a different image degradation model. Targeting image enhancement for degraded images but not being limited to it, the proposed method provides a highly effective solution for both color and gray images with substantial improvement in the process time compared to conventional methods. The proposed method introduces a hybrid scheme of two image enhancement modules: a model‐based (extensive) enhancement and a model‐free (immediate) enhancement. The proposed method is validated by using simulated synthetic color images and real‐world color and grayscale underwater images. Real‐world validation is performed in various environments such as hazy indoor, smoky indoor, and underwater. Using the ground truth trajectory or clear images acquired from the same area but without turbidity, we evaluate the proposed visibility enhancement and camera registration improvement for a feature based (ORB‐SLAM2), a direct (LSD‐SLAM), and a visual underwater SLAM application.

show abstract

Estimation of ambient light and transmission map with common convolutional architecture

Cited by 63 publications

References 17 publications

Generation and Processing of Simulated Underwater Images for Infrastructure Visual Inspection with UUVs

Generation and Processing of Simulated Underwater Images for Infrastructure Visual Inspection with UUVs

Underwater Image Restoration Based on a Parallel Convolutional Neural Network

Channel invariant online visibility enhancement for visual SLAM in a turbid environment

Contact Info

Product

Resources

About