Comparative Analysis of Threshold Based, K-means and Level Set Segmentation Algorithms

Image segmentation algorithms provide a mean to segment the images of different formats. The beauty of an algorithm lies within the quality of segmentation it performs. The large number of segmentation algorithms are available, analysis of algorithms is emerging as an interesting field. A number of algorithms are available for segmenting an image but there is not a single algorithm which performs better for all applications. In this paper, the comparative analysis of level set, K-means and Thresholding based algorithms has been presented for standard Lena and Cameraman image. The performance of the algorithms measured in form of white pixel ratio and processing time. Over the last few years image processing has been playing an important role in the analysis of various types of images acquired both in visible band of Electromagnetic (EM) spectrum and remaining of the EM spectrum band. Finding out the exact position of the object in the image requires segmentation. Numerous segmentation and algorithms have been proposed for this purpose. Consequently there was a need for the comparison of different segmentation algorithms based on some parameter instead of taking into account the perceptual quality of the segmented image, which would take into account some other attribute of the segmentation algorithm. The algorithms taken for comparative study are Level set algorithm, K-means algorithm and Thresholding based algorithm. For the judgment of quality of a segmented image several parameters have been propounded by different scholars. These methods evaluate the performance of the segmented image and provide a way for the selection of an optimal technique for image segmentation. Now a day, there are cutting edge algorithms available which provide a very good segmentation of the images. So some parameter, different from the perceptual quality of the segmented image, has to be taken into account to evaluate segmentation algorithms. The task of evaluation becomes difficult when the performance matches to a great extent. So some parameter which compares the performance of the segmentation algorithms in terms of processing time is an interesting idea. Everything in this universe is bounded by time and space. In the context of segmentation algorithms, time may be considered as an important parameter to be taken into account. Here time may be regarded as a parameter in the sense that different algorithms take different time to segment the same image. We may call this parameter as processing time. Apart from processing time, another parameter taken into account for the comparative analysis of different algorithms is White Pixel Ratio (WPR). This ratio indicates the proportion of white pixels within a segmented image. The paper is organized as follows. Section II, describes level set, K-means and Thresholding based algorithms. Section III, deals with segmentation evaluation parameters. Section IV, discusses the related work. Experimental results are discussed in Section V, while Section VI, expounds conclusion. II. Different Algorithms Taken for Comparison