Image And Video Dehazing Based On Improved Atmospheric Light Scattering Model And Bounded Transmission
Poor weather conditions, such as haze and rain, obscure images and videos and degrade their quality. This degradation is due to the presence of atmospheric particles that scatter and attenuate light. This effect threatens the reliability of many outdoor applications such as driver assistance. Dehaz...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | en_US |
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Poor weather conditions, such as haze and rain, obscure images and videos and degrade their quality. This degradation is due to the presence of atmospheric particles
that scatter and attenuate light. This effect threatens the reliability of many outdoor applications such as driver assistance. Dehazing, a technique to restore poor-quality images and videos is thus vital in image enhancement. However, dehazing is highly challenging due to its ill-posed nature when considering only hazy images as inputs.
To solve the dehazing problem, many methods attempted to overcome this ambiguity
employing atmospheric light scattering (ALS) model. To restore the original scene colors using ALS, two variables have to be estimated; transmission which is the original irradiance of scene objects and airlight which is the ambient light. However these methods are too slow to be considered in real-time applications due to the nonlinear transmission estimation and refinement or achieve poor dehazing quality. Another problem is that the application of single-image dehazing methods to videos results in flickering artifacts. These artifacts are caused by the lack of inter-frame spatiotemporal coherence. This thesis proposed a scheme that consists of two phases to solve single-image and video dehazing problem. In phase one, a single-image dehazing method called intensity deterioration ratio and saturation deterioration ratio (IDRSDR) is first proposed, IDRSDR is able to achieve real-time dehazing of high quality images due to its straight-forward transmission estimation. Second, considering that speed-quality tradeoff is inevitable in real-time practical dehazing applications, another method is proposed to enhance the quality of restored images called bounded transmission (BT). Phase two proposed a smoothing edge-preserving filter, namely, controlled Gaussian filter (CGF) to refine the transmission and ensure spatial coherency of dehazed videos. Combining the two phases in one scheme real time video dehazing is achieved (>=24 fps). Objective and subjective evaluations were conducted on three image datasets and one video dataset. Testing results have demonstrated that IDRSDR is faster than existing methods 5 to 10 times with statistically similar quality. In addition, the findings indicate that BT is at least 2 times faster than existing methods with statistically similar or better quality. Furthermore, results of phase two proved that employing CGF as a transmission refinement filter enabled the real-time video dehazing while maintaining higher degree of naturalness than that of the existing filters; CGF improved the speed of video dehazing by 250% to 530% with at least 2% better quality. Tests also indicated that coupling CGF with IDRSDR in one scheme is feasible to be used in real-time single-image dehazing, whereas coupling CGF with BT is more suitable to be used in real-time video dehazing, especially when the scene is expected be dynamic, such as in driver assistance applications. |
|---|