Login | Register

Image Segmentation and Adaptive Contrast Enhancement for Haze Removal

Title:

Image Segmentation and Adaptive Contrast Enhancement for Haze Removal

Bao, Zhu (2021) Image Segmentation and Adaptive Contrast Enhancement for Haze Removal. Masters thesis, Concordia University.

[thumbnail of Zhu_MASc_F2021.pdf]
Preview
Text (application/pdf)
Zhu_MASc_F2021.pdf - Accepted Version
14MB

Abstract

Nowadays, hazing scenes are very frequent in images acquired outdoors. For such images to be used as input images of autonomous systems, it is important to restore the image details so that they can provide sufficient information to the system. As hazy images feature poor contrast due to degraded image variations, one can use a contrast enhancement method, such as CLAHE, to restore the image details. However, in case of very heavily hazy images, the image signal quality is severely degraded. Applying a strong enhancement may help to recover the details but will meanwhile generate very visible noise, affecting the image quality. In order to handle the problem of the conflict of the degree of enhancement and noise created in the process, it is thus necessary to develop a good algorithm with different enhancement based on a specific mask and signal variations in different regions.
In this thesis, a novel dehazing algorithm is proposed, which aims at heavily hazy images. In order to restore effectively image details that are almost invisible in hazy images without over enhancement in other foreground areas, the proposed algorithm involves a new adaptive CLAHE process, in which a stronger enhancement is applied to the areas of weaker variations, different from an existing version of improved CLAHE with adaptive clip limit. This new CLAHE is applied only to the foreground areas, by means of a protective mask, so that there will not be noise enhancement in the atmospheric background and the other flat areas. Each input image is segmented into foreground and background areas to generate the mask. In case of heavily hazy images, the gradient amplitudes of the signals and the noise are in the same level and it is thus very difficult to distinguish foreground and background areas. A new gradient matrix has been defined and a gradient feature vector proposed to detect the locally dominant gray level variations, with a view to identifying the pixels of very weak variations in foreground areas with the noise presence. This gradient vector helps to distinguish the foreground and background areas in heavily hazy images, and the segmentation can be done effectively, which makes it possible to apply the new adaptive CLAHE without noise enhancement.
The proposed algorithm has been tested with different kinds of hazy images. In case of heavily hazy image input, it performs better, in terms of image detail restoration, than existing methods based on dark-channel-prior (DCP) or other form of CLAHE. It is effective with hazy images of high dynamic range. It is also useful in case of lightly hazy images.

Divisions:Concordia University > Gina Cody School of Engineering and Computer Science > Electrical and Computer Engineering
Item Type:Thesis (Masters)
Authors:Bao, Zhu
Institution:Concordia University
Degree Name:M.A. Sc.
Program:Electrical and Computer Engineering
Date:November 2021
Thesis Supervisor(s):Chunyan, Wang
ID Code:989961
Deposited By: Bao Zhu
Deposited On:16 Jun 2022 15:21
Last Modified:16 Jun 2022 15:21
All items in Spectrum are protected by copyright, with all rights reserved. The use of items is governed by Spectrum's terms of access.

Repository Staff Only: item control page

Downloads per month over past year

Research related to the current document (at the CORE website)
- Research related to the current document (at the CORE website)
Back to top Back to top