M.Sc. Tezi Görüntüleme | |||||||||||||||||||||
|
|
||||||||||||||||||||
Summary: Today, with the development of technology, digital images are used in many environments. Accordingly, many methods have been developed to process these digital images and to use them for various purposes. One of the first and the most important steps in image processing is segmentation. Through segmentation, semantic data is obtained for the next steps of the image processing. However, due to the type of data used, the intent of the application, and many others, a segmentation method does not always give desired success. For these reasons, various segmentation methods are proposed for different problems. One of the most remarkable among the interactive segmentation methods is the GrabCut interactive segmentation method which is developed by utilizing graph structure. In this thesis, the shortcomings and weaknesses of GrabCut interactive segmentation method were studied and novelties in the literature for this method were analyzed. Efficient pre-processing methods that do not modify the original structure of the GrabCut method have been tested and the results were compared with the original GrabCut method. As a result of the experiments, it was seen that, the conversion of the input color space to the CIE-based color spaces and the use of contrast enhancement methods (CLAHE and Perception-Based Contrast Enhancement of Images) as a pre-process contributed positively to the success. There is no improvement in the use of noise reduction methods on the noisy images.
Key Words: Segmentation, GrabCut, User Interactive Segmentation, Graph Structure. |