الفهرس 
Literature Survey for Video Dehazing AlgorithmsOnly 14 pages are availabe for public view
 |  | from | 167 |  |  |
| | | from | 167 | | |
Abstract
This thesis is concerned with video quality enhancement using dehazing
techniques. Dehazing is the process of removing haze from hazy images and
enhancing the image contrast. Hazing is defined as a degradation effect in
images. It is produced due to light scattering from tiny particles in the
atmosphere combined with the sunlight around the scene to be imaged.
Hazing resembles a white mask overall image. Most automatic systems,
which depend on the description of the input images, suffer from the worst
results due to degraded images. To enhance video quality, we use three types
of dehazing algorithms in this thesis such as an optimized dehazing
algorithm, a recursive deep residual learning (DRL) network, and a dual
transmission map-dehazing algorithm.
Our proposed algorithm depends on pre-processing of frames before
dehazing process to remove noise and reduce dynamic range because all
frames contain some noise, and limited dynamic range due to sensor
measurement error that can be amplified in the haze removal process if
ignored. We use different types of enhancement techniques to remove noise
and reduce dynamic range before the dehazing process such as median,
homomorphic enhancement, and Frost filter. The video consists of several
frames (images). We work on two types of video sequences: a Near Infrared
(NIR) and a visible sequence. Results for visible video are more obvious
than NIR video because hazy visible frames have a large amount of hazing
compared to hazy NIR images. The effect of attenuation parameter of
atmosphere α and the effect of attenuation weight ω without and with
enhancement visible and NIR; video frames are investigated in our proposed with an optimized dehazing algorithm.
We modify the deep residual learning algorithm to generate a new
one to be suitable for both NIR and visible frames. The number of iterations
in the DRL network is increased from three to nine iterations to study the
effect of increasing the number of iterations on the output-dehazed frames.
We stopped at nine iterations because elapsed time increases with increasing
the iteration number. The effect of a regularization parameter β on visible
and NIR dehazed frames without and with homomorphic enhancement is
investigated in our proposed dual transmission map-dehazing algorithm. The
Peak Signal-to-Noise Ratio (PSNR) and correlation after the dehazing
process (between dehazed and input hazy frames) and the spectral entropy of
dehazed frames are used as metrics methods for our proposed algorithm.
Comparison between the proposed algorithm and different dehazing
algorithms on real-world images is achieved. The simulation results prove
that enhancement for average results of five frames for visible
(riverside.avi), video is 14.74% by dual transmission dehazing technique
followed by optimized dehazing by 9.65%. Finally, DRL Network by 9%.
For NIR video, the best enhancement by optimized dehazing with 22.28%,
after that dual transmission dehazing algorithm by 21.57%. Finally, DRL
network by 4.6%.