Comparative Analysis of Image Segmentation Techniques

INTRODUCTION

The recent applications of Digital Image Processing are used in medicine, photography, remote sensing film, video production, and security monitoring [1]. Many modern technologies are emerged in the fields of Image Processing, especially in Image Segmentation domain. Generally, Segmentation is the process of segmenting or partitioning a digital image into multiple segments or partitions .These segmented partitions are analyzed and processed to get some meaning images, then cluster those image pixels into prominent image regions, i.e., regions corresponding to individual objects, surfaces or grouped parts of objects. Image segmentation algorithms are based mainly on two properties (i.e.) either discontinuity principle or similarity principle. The idea behind the discontinuity principle is to extract regions that differ in properties such as intensity, color, texture, or any other image statistics and the similarity principle is to group pixels based on common properties in Image Segmentation [2].Image Segmentation widely used in Face Recognition, Medical Field, Astromical and many other fields. Bali discussed about principle segmentation techniques, implementation, and applications based on human and machine perceptions [3].

KEY TECHNIQUES IN IMAGE SEGMENTATION

Segmentation Algorithms have been developed to segment the images and it can be classified into following  Segmentation by Clustering  Segmentation by Edge Detection  Segmentation by Fuzzy Logic  Segmentation by Neural Network  Segmentation by Region Based  Segmentation by Thresholding

Segmentation by Clustering

Clustering is the task of dividing the population (data points) into a number of groups, such that data points in the same groups are more similar to other data points in that same group than those in other groups. These groups are known as clusters. The most commonly used clustering algorithm is k-means. The loss of information image is often due to generation of boundary in expression of images during segmentation. The clustering method is applied to classification research of many studies [4].K-means algorithm for Image Segmentation helps to improve high performance and efficiency. This method works based on flow as shown in Figure 1. In addition, selection is based on number of clusters determined using datasets from images by using frame size and the absolute value between the means of clusters. The experimental result provides better output and increases the speed of the execution process. K-means used to estimate the number of cluster that dependent on the values of pixels. The number of iteration of process affects its computational cost routine for reaching convergence. The computational time as well as segmentation quality aspects helps to improve accuracy and implementation time by using K-means Clustering [5].

Figure 1

The General Structure of K-means Clustering Based Image Segmentation

Segmentation by Edge Detection

In Edge Detection, Segmentation is done from end to end by identifying the boundaries; edges are detected to identify the discontinuities in the image. In Edge based Detection, detected edges need not to be closed. By detecting edges or pixels during segmentation helps us to extract or associate to form a closed object boundaries [7]. The process of identifying and locating sharp with fine discontinuities is called Edge Detection and this technique helps to get the desired output [8].Performance of various edge detection techniques is carried out with traditional ones as shown in Figure 2. An improved Canny Algorithm is performed and tested by comparing with various Edge Detection Algorithms and concluded that Canny is good one. Canny method helps to separate noise from image before finding edges in image. [9].Edge Detection techniques help to retain change from grey tones in image. means strategies can improve remote sensing image using Threshold Segmentation with fewer iterations time, good stability and robustness [11]. This method mainly depends on the mean of each bunch and gathering similar information characteristics into one group. Fuzzy Cognitive Maps (FCM) is one of the reasonable clustering techniques in Medical Image Segmentation mainly for Magnetic Resonance Imaging (MRI). Fuzzy C-mean Algorithm gives the better results in image segmentation [12].The general structure of Fuzzy logic algorithm is shown in Figure 3. The Deep Convolution Neural Network helps to learn the hierarchy features (low to mid to high level). Deeper means better feature extraction and need to regularize the model well and finely tuned to network that can reduce the domain mismatch [14].

Figure 2

The General Structure of Edge Based Image Segmentation

Figure 3

The General Structure of Fuzzy Based Image Segmentation

Segmentation by Region Based

Region Based Segmentation is also called as Similarity Based Segmentation. As shown in Figure 5 an RGB image which is given as input is converted into gray image to perform segmentation by using Morphological Operations.

Figure 5

The General Structure of Region Based Image Segmentation

One of the important Morphological technique is Watershed

Algorithm. Based on this algorithm, image is viewed in topographic surface and its gray levels of a pixel are interpreted by its altitude. Suppose a water source is placed in each Regional Minimum (also called `Catchment Basins') and the entire topography structure is flooded below in surface. When water from two sources (regional minima) are about to meet, a dam is constructed to prevent the merging .The flooding and dam construction process continues until only the dams are visible from above. These dams (Watershed lines or 'Watersheds') effectively segment the image into regions. Many morphological techniques are available for the segmentation of images. But the problem related to that segmentation is condensed or defeated by the better selection of Marker Controlled Algorithm. Marker can be applied directly on gradient image to control over segmentation [15]. Different Segmentation techniques are reviewed and found that instead of segmenting regions, marking is very easy [16]. Watershed Algorithm is a powerful and efficient in case of overlapping or adjacent rocks formation. By combining with the process of mark selection using Field-Programmable Gate Array (FPGA) Processor, Segmentation can be done effectively [17].

Segmentation by Thresholding

The process of Image Segmentation provides the partition of image into different segment according to their feature attribute. Thresholding is the simplest method of Image Segmentation. The local thresholding technique used Region Based Segmentation process and used multiple thresholds for the process of segmentation. The pixel values falling below or above that threshold can be classified accordingly (as an object or the background) [18].This technique is known as Threshold Segmentation. Various algorithms like Otsu's, Eridas and Quadratic Integral Ratio (QIR) are used to do segmentation as shown in Figure 6. One of the Statiscal Algorithm is Otsu's for finding histogram of gray image with Statiscal distribution gives good result [19].The image threshold based on gray level has been calculated and pixels are calibrated to obtain best segmentated image [20]. IJSTR©2019 www.ijstr.org

Figure 6

The General Structure of Thresholding Based Image Segmentation

COMPARISION OF SEGMENTATION TECHNIQUES

Comparison of Image Segmentation is made and its advantages along with limitations are discussed and tabulated.

RESULTS AND DISCUSSIONS:

The image segmentation results using Marker-Controlled are shown in Figure 7. Figure 7(a), 7(b), 7(c) shows original, Gradient, Watershed Transformed Gradient respectively. Then it is found that over segmentation arises which can be reduced with implementation of Marker-Controlled Watershed Segmentation as shown in Figure 8.

Figure 7

(a) Read the Image Figure 7: (b) Gradient Magnitude Image ( c) Watershed Transformed Gradient Magnitudes Image ISSN 2277-8616 2210 IJSTR©2019 www.ijstr.org

Figure 8

(g) Modified Regional Maxima Figure 8 :( h) Threshold Opening -closing Superimposed on Original Image by Reconstruction ( i) Watershed Ridge lines Figure 8 :( j) Markers and object Boundaries Superimposed on Original Image ( k) Colored Watershed Label Matrix Figure 8 :( l) Colored Labels Superimposed Transparently on Original Image 4. CONLUSION Several important techniques for Image Segmentation are reviewed and summarized. Algorithms like K-means, Canny Edge Detection, Fuzzy C-means, Neural Network, Morphological Watershed, Otsu's Thresholding techniques are discussed and compared. Some of the recent works on Image Segmentation are analyzed. With the subsequent observation of these techniques individually, it is concluded that,

Watershed Segmentation using the Distance Transforms

The existing edge detecting operators such as Prewitt, Robert, Canny and LoG helps to segment image by using Distance Transform in Morphological Watershed Segmentation. The traditional Watershed Segmentation using Distance Transform are applied to detect edges which are more apparent, pinpointed and sharp with abundant edge information. Moreover, Canny Edge operator filers the noise more effectively than Sobel, LoG and other traditional operators in Watershed Segmentation using Distance Transform. It is found that Canny Edge operator is more efficient for edge detection with watershed segmentation using distance transform method [21].The steps to be followed includes

Step 1: Read the Image Step 2: Convert it to Binary

Step 3: Find complement of Binary image and apply distance transform.

Watershed Segmentation using Gradients

The preprocessing of a gray-scale image before using the Watershed transformation for segmentation can be done with the help of gradient magnitude. Dilation and erosion can be used in combination with image subtraction to obtain the Morphological Gradient image with the smoothened image. The regions in an image are thickened and shrunk by dilation and erosion. Watershed Segmentation based on Morphological Gradient are introduced in Watershed Segmentation through opening and closing by reconstruction. Then reconstruction operators are in use to restructure gradient image in which a set of gradient pixels with high value are conserved and few gradient pixels with low value are detached. Thus improved algorithm using Gradients is applied to reconstruct image which eliminates over-segmentation but not completely. Although it holds the position of region contours clearly [22]. The steps to be followed includes

Watershed Segmentation using Marker-Controlled

From the above discussion it is found that Watershed Segmentation applied directly on Gradient magnitude images, it shows the problem of over segmentation. In Marker-Controlled Watershed Segmentation, the Image to be segmentated is converted in to Grayscale. Then reconstruction of image is made by opening and closing by reconstruction with selection of markers based on foreground and background objects, to found the definite boundaries. The accuracy of the Marker-based Watershed Segmentation is high compared to Sobel, Roberts, LoG, Canny, Active contour Morphological-based Segmentation, Otsu's thresholding segmentation methods [23].It is reliable, efficient, robust, and also provides result with reduced noise .It also works fine with composite images. The steps to be followed includes Marker-Controlled Morphological Watershed Segmentation is superior. Over segmentation is major drawback in segmentation, Marker-Controlled Watershed Segmentation helps to solve and bring out segmentation with reduced over segmentation. It brings desired output when applied using distance measure and gradient magnitude operations. Since there is no unanimously accepted technique for Image Segmentation process, a good recommendation is that Marker-Controlled Watershed Segmentation Algorithm is best suitable for segmentation to attain better implementation in Medical Image Processing.