Research Information System
Thesis Type: Doctorate
Institution Of The Thesis: Gazi Üniversitesi, Fen Bilimleri Enstitüsü, Turkey
Approval Date: 2016
Student: ORHAN EMRE ÇELİKNALÇA
Co-Supervisor: ÇETİN ELMAS, RECEP DEMİRCİ
Open Archive Collection: AVESIS Open Access Collection
Abstract:Digital image processing includes image restoration, compression, segmentation and object recognition in a computer environment. Image segmentation is a classification of pixels in image in order to create meaningful regions and process which enables to recognize objects in image, accordingly. Thresholding, edge detection, region-based and classification techniques are the most common methods which are used for the relevant process. Classification technique is based on grouping the similar pixels in an image. However, the greatest drawbacks of conventional methods are that they require proper selection of threshold and of cluster numbers and therefore human intervention. Additionally, there is no consistent method to choose the cluster centers and also they are slow due to their iterative nature. In this study, a new classification algorithm based on three-dimensional histograms of grayscale and color images has been proposed. Peaks of histogram are assigned to be cluster centers and pixels that are close or similar to these peak pixels are assigned to the relevant cluster. According to the findings of psychologists working on the science of perception, human perception represents similar characteristics of Gaussian function that depends on the Euclidian distance. Therefore, in order to estimate the color similarity of pixels, Euclidean distance and Gaussian function on the color space was used. During the classification process, similarity threshold value has been adaptively estimated and, hence human intervention has been eliminated. Moreover, number of clusters has been automatically detected by analyzing the histograms. A user interface has been developed and, comparative results with conventional methods have been obtained with the proposed algorithm. It was observed that the proposed method produced results faster than conventional methods and more compatible with the human perception.