FÜZYON TABANLI HİBRİT SİS KALDIRMA


Thesis Type: Postgraduate

Institution Of The Thesis: Bursa Uludağ University, FEN BİLİMLERİ ENSTİTÜSÜ, FEN BİLİMLERİ ENSTİTÜSÜ, Turkey

Approval Date: 2023

Thesis Language: Turkish

Student: BAHADIR ARABALI

Supervisor: Kemal Fidanboylu

Abstract:

Digital images obtained outdoors may cause the deterioration or loss of important details, objects and colors due to the fog formed in the atmosphere. Correcting the fog-related distortions in the images, revealing the areas of interest and removing the fog from the image is called haze removal. The filter-based Dark Channel Prior (DCP) algorithm developed for this purpose has an important place among fog removal algorithms. DCP algorithm is based on the observation that fog-free outdoor scene images have very low pixel density values in different color channels. With this algorithm, a patch is created for the fog in the image and the fog in the image can be removed or reduced. One of the important and widely used example of image fusion algorithms is the Exposure Fusion algorithm. This algorithm weights the image captured at different exposure levels according to the saturation and contrast values and corrects the exposure level of the image with the fusion process it performs. Within the scope of this thesis, the process of removing fog from the image was performed more successfully by using the Dark Channel Prior and Exposure Fusion algorithms together. Thus, images that are less structurally distorted, more natural and areas of interest free from fog images were obtained after defog removal. The developed approach consists of firstly correcting the exposure of the image containing the fog, and then taking the exposure corrected image into the fog removal process with the DCP algorithm. In this way, the fog removal performance of the DCP algorithm has been increased. The proposed method has been tested with images in the O-Haze database, which consists of fog images, and the results are compared with different performance metrics.