<script async src="//pagead2.googlesyndication.com/pagead/js/adsbygoogle.js"></script>
<!-- martinchinwe -->
<ins class="adsbygoogle"
style="display:inline-block;width:728px;height:90px"
data-ad-client="ca-pub-3432567292388333"
data-ad-slot="3193770573"></ins>
<script>
(adsbygoogle = window.adsbygoogle || []).push({});
</script>
2.1.Image segmentation
Image segmentation, is the process of separating the object (foreground) from the background, as well as dividing the images into separate regions, each region having similar characteristics [75]. The accuracy of segmentation technique used determines the success or failure of computerized digital image analysis procedures. It also involves counting the number of components in the image, pointing out their locations and extracting the component contour. Segmentation assigns labels to every pixel in an image such that pixels with the same label share some virtual characteristics [11]. The purpose of segmentation is to simplify the image to make it easy for analysis since subsequent techniques for image description and recognition depend on the image segmentation. The image segmentation has many applications in medicine such as: visualization and volume estimation, organs abnormality detection and tissue quantification.
The segmentation techniques are based on either of the two basic properties of intensity values: Discontinuity techniques and Similarity techniques [76]. The property of discontinuity technique partitions, images based on the abrupt changes in the intensity value of each pixel, like the edge in an image. On the other hand, the similarity techniques are based in partitioning of the image into regions that are similar according to established condition. The thresholding technique is based on the similarity properties of thresholding. There are many methods of image segmentation techniques: thresholding, watershed.
2.1.1. Thresholding technique
The thresholding technique is one of the most efficient and simplest segmentation method of separating foreground of an image from the background [77][76] using appropriate threshold (T). Based on the method of selecting T, two types of thresholding methods exist: the local thresholding technique and global thresholding technique [78].
Fig. 2.15: Selection of threshold using image histogram
The global technique which is used when T is constant fails when the background illumination is not uniform. When the background is unevenly illuminated, local technique multiple thresholds are used to recursively compute different values of T to compensate for uneven illumination. The thresholding technique is a non-linear operation that converts grayscale image into a binary image [11] using threshold value (T). The threshold, T for a particular image depends on frequency of occurrence of the pixel intensities such that the binary image contains all the essential information about the position and shape of the image of interest. The most common way of converting grayscale image to the binary image, is to select a single threshold value such that for below or equal to will be classified as black (0.0) and for above T, will be classified as white (1.0). The binarization of an image using the concept of thresholding according to [45] is shown in equation (2.50). The most common problem with thresholding technique is selecting the appropriate value of the threshold (T). It is to be noted that for grayscale images, and frequently used method of selecting the threshold is the histogram technique.
The ideal case is when the histogram presents only two dominant modes and outstanding valley (bimodal) in which the threshold exists at the valley point. In practice, the histogram techniques are more complex when there are many peaks and no clear-cut valley.
For more accurate threshold selection, the Otsu's method invented by Nobuyuki Otsu is used [77]. It is used to automatically convert the grayscale image to a binary image. The Otsu's method was used extensively in the work done by [45][77]. The algorithm assumes that the image to be segmented contains two classes of pixels (foreground and background) and it calculates the optimum threshold separating the two classes such that the intra-class variance is minimal.
In Otsu method, a threshold value (T) that minimizes the intra-class variance is recursively found using expression according to [45]. Otsu shows that minimizing the intra-class variance is the same as maximizing inter-class variance . The inter-class variances are defined as the weighted sum of variances of the two classes, the weighted probability.
For effective determination of threshold using the Otsu method, the operation is carried out iteratively. This is done by computing the values of at various values of T. The value of T in which the has the maximum value is the Threshold. Often time, two values of T corresponding to the two maximum values of are used to compute the threshold.
The segmented image obtained from thresholding has the advantages of having smaller storage size, easy to process and manipulate. However, thresholding technique is limited by the facts that it uses only two classes and therefore does not apply to multichannel image (colour image and multispectral image). In addition, it does not take into account the spatial characteristics of an image due noise.
2.1.2. Watershed segmentation approach
The watershed segmentation method is most suited when there is a need to separate touching objects within an image. The technique was extensively used in the work done by [75]. The separation of touching objects is a very tough task and not all the segmentation technique can handle the operation. The watershed method works by connecting blobs of pixels in the foreground and then the blob of pixels in the background separately, such that the maxima occur only in the background and foreground. The method works best when the foreground and the background of the image to be segmented can be identified. The watershed technique follows these basic steps:
· Compute a segmentation function. This is an image whose dark regions are the objects to be segmented.
· Compute foreground markers. These are connected blobs of pixels within each of the objects.
· Compute background markers. These are pixels that are not part of any object.
· Modify the segmentation function so that it only has minima at the foreground and background marker locations.
· Compute the watershed transform of the modified segmentation function.
The watershed method applies both the edge detection and region growing in image segmentation.
The basic concept of the watershed is to separate a gray level image into three representations such as: minima (region of intensity close to 0.0), catchment basins (regions of intensity close to 1.0) and watershed, according to [75] as illustrated in Fig.2.16 and the aim of the watershed is to find the region of high intensity that divides neighbouring local minima [75]. The watershed method produces good results for gray level images with different minima and catchment basin and for binary image with only two levels (1 and 0); the catchment basins correspond to the homogeneous gray level regions of the image. The method gives good results when no two black pixels connected. However, when two black pixels are connected, it is necessary to preprocess the image using distance transform (DT). The DT of a binary image is the distance between two closest pixels of opposite colour. The commonly used DT is the Euclidean distance (d). The Euclidean distance according to [75]. The watershed works by extracting the seeds that indicates the presence of objects at a specific location in the image. The method uses the concept of gradient to group the image regions of high intensity gradient that divide local minima. Unlike, other classical segmentation methods, the watershed method operates on the gradient image.
The gradient image is defined as the first partial derivative of an image and contains the information about all the measurements for the change of gray level. The gradient values, from the initial segmented image are obtained using gradient operator according to [79]. The gradient image values are used to calculate the edge strength values of the input image.
The watershed method embodies the other major image processing techniques such as discontinuity detection, thresholding and region processing and because of these factors; watershed method is more effective and stable. In practice, this transform leads to over-segmentation due to noise or local irregularities in the gradient image, but over-segmentation can be overcome by merging method based on mean gray values and edge strengths. However, the best method of overcoming over-segmentation is to filter the image before segmentation.