Single image desmogging using oblique gradient profile prior and variational minimization
- PDF / 2,157,386 Bytes
- 17 Pages / 439.37 x 666.142 pts Page_size
- 27 Downloads / 179 Views
Single image desmogging using oblique gradient profile prior and variational minimization Jeevan Bala1 · Kamlesh Lakhwani1 Received: 21 August 2019 / Revised: 16 December 2019 / Accepted: 5 February 2020 © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract An efficient estimation of transmission map for desmogging model is an ill-posed problem. The quality of restored image depends upon the accurate estimation of transmission map. However, transmission map obtained using various dehazing models is not accurate in case of images with large haze gradient, and fail while image desmogging. As a result, the restored images suffer from numerous issues such as halo and gradient reversal artefacts, edge and texture distortion, color distortion, etc. Therefore, this paper designs a novel transmission map estimation by using weighted integrated transmission maps obtained from foreground and sky regions. Additionally, transmission map is further refined using an integrated variational regularized model with hybrid constraints. However, the proposed technique suffers from hyper-parameters tuning issue, therefore, in this paper, a non-dominated sorting genetic algorithm is also used to tune the hyper-parameters of the proposed technique. The comparison of designed desmogging model is also done with other dehazing models by considering benchmark and real-time hazy images. The comparative analyses reveal that the designed model outperforms existing models subjectively and quantitatively. Keywords Desmogging · Gradient channel · Restoration model · Transmission map · Variational minimization
1 Introduction Images obtained in poor environmental circumstances such as haze, fog, smog, etc. suffer from poor visibility issue. The optical imaging model is formulated as a linear combination of an actual scene radiance, airlight and the transmission map. It is mathematically defined as (He et al. 2011): α (δ) = κ (δ) μ (δ) + (1 − μ (δ)) ν, (1)
B
Jeevan Bala [email protected] Kamlesh Lakhwani [email protected]
1
Computer Science and Engineering Department, Lovely Professional University, Jalandhar, Punjab, India
123
Multidimensional Systems and Signal Processing
Here, κ denotes the actual object radiance and α represents the obtained hazy image. ν and μ show the global atmospheric light and transmission map, respectively.The foremost function of dehazing method is to restore κ from α. Though, atmospheric light (ν) and transmission (μ) are unknown. To evaluate the atmospheric light (ν) and transmission (μ), many dehazing models have been designed so far. Many authors have designed multiple-images based dehazing models (Liu et al. 2016). These models demand additional information of input images in prior (Riaz et al. 2016; Narasimhan and Nayar 2003; Nayar and Narasimhan 1999). However, in real-time dehazing it is hard to obtain additional information of the given scene (Li et al. 2014; Wang and Zhu 2015). Guo et al. (2017) implemented a fusion based dehazing model. Yoon (2016) designed an adaptiv
Data Loading...
