Imaging-Model-Based Visibility Recovery for Single Hazy Images
- Publication Type:
- Thesis
- Issue Date:
- 2020
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Low-quality images captured in hazy weather can seriously impair the proper functioning of vision system. Although many meaningful works have been done to realize the haze removal, there are still two key issues remain unsolved. The first one is the long processing time attributed to the involved tools; the second one is existing prior employed in state-of-the-art approaches cannot be suitable for all situations. To address such problems, a series of haze removal techniques have been developed. The main contributions of this dissertation can be summarized as the following.
For efficiency, a gamma correction prior is proposed, which can be used to synthesize a homogeneous virtual transformation for an input. Relying this prior and atmospheric scattering model (ASM), a fast image dehazing method called IDGCP is developed, which converts single image haze removal into multiple images haze removal task.
Unlike the IDGCP, another solution for accelerating dehazing (VROHI) is to utilize a low complexity model, i.e., the additive haze model (AHM), to simulate the hazy image. AHM is used on remote sensing data restoration, thus the first step of VROHI is to modify the AHM to make it suitable for outdoor images. The modified AHM enables to achieve single image dehazing by finding two constants related to haze thickness.
To overcome the uneven illumination issue, the atmospheric light in ASM is replaced or redefined as a scene incident light, leading to a scene-based ASM (Sb-ASM). Based on this Sb-ASM, an effective image dehazing technique named IDSL is proposed by using a supervised learning strategy. In IDSL, the transmission estimation is simplified to simple calculation on three components by constructing a lineal model for estimating the transmission.
According to previous Sb-ASM and the fact that inhomogeneous atmosphere phenomenon does exist in real world, a pixel-based ASM (Pb-ASM) is redefined to handle the inhomogeneous haze issue. Benefitting from this Pb-ASM, a single image dehazing algorithm called BDPK that uses Bayesian theory is developed. In BDPK, single image dehazing problem is transformed into a maximum a-posteriori probability one.
To achieve high efficiency and high quality dehazing for remote sensing (RS) data, an exponent-form ASM (Ef-ASM) is proposed by using equivalence infinitesimal theorem. By imposing the bright channel prior and dark channel prior on Ef-ASM, scene albedo restoration formula (SARF) used for RGB-channel RS image is deduced. Based on Rayleighąŕs law, SARF can be expanded to achieve haze removal for multi-spectral RS data.
Please use this identifier to cite or link to this item: