Image Segmentation Using Split and Merge Techniques with a Mixture of Heterogeneous Distributions

Abstract

Image segmentation is a technique that partitioned the image into prerequisite semantic unique regions. Simplifying the representation of an image into something more easily to analyze and meaningful is the ultimate goal of segmentation. It is used for locating boundaries and objects in an image such as lines, curves or object. Segmentation serves many computer applications such as pattern recognition, object recognition, automatic traffic control, and many other applications. Image segmentation is considered the bottleneck in many image processing techniques. There are a vast number of segmentation techniques that is available but none of them satisfy the global properties, so it is remain challenge for researcher to find best one. Thresholding is one of the simplest and effective techniques for image segmentation. It defines a threshold value T, and the gray values less than T will be considered as a class, and those above T belong to another class. The estimation of the optimal threshold T is still the big problem for image thresholding. The aim of this thesis is to introduce a new method for image thresholding based on histogram split-merge technique with a mixture of heterogeneous distributions. The proposed method will be applied on different type of synthetic and real images; the results obtained will be compared to those of the existing methods.