Semiautomatic methods can partially remove the subjectivity in defining ROIs by human operators. The most commonly used method is by means of thresholding because of its simplicity in implementation and intuitive properties. In this technique, a predefined value (threshold) is selected and an image is divided into groups of pixels having values greater than or equal to the threshold and groups of pixels with values less than the threshold. The most intuitive approach is global thresholding, which is best suited for bimodal image. When only a single threshold is selected for a given image, the thresholding is global to the entire image. For example, let f (x, y) be an image with maximum pixel value Imax, and suppose r denotes the percent threshold of the maximum pixel value above which the pixels will be selected, then pixels with value p given by r

can be grouped and a binary image g(x, y) is formed:

I 0 otherwise in which pixels with value of 1 correspond to the ROI, while pixels with value 0 correspond to the background.

Global thresholding is simple and computationally fast. It performs well if the images contain objects with homogeneous intensity or the contrast between the objects and the background is high. However, it may not lead itself fully automated and may fail when two or more tissue structures have overlapping intensity levels. The accuracy of the ROI is also questionable because it is separated from the data based on a single threshold value which may be subject to very large statistical fluctuations. With the increasing number of regions or noise levels, or when the contrast of the image is low, threshold selection will become more difficult.

Apart from global thresholding, there are several thresholding methods which can be classified as local thresholding and dynamic thresholding. These techniques maybe useful when a thresholding value cannot be determined from a histogram for the entire image or a single threshold cannot give good segmentation results. Local threshold can be determined by using the local statistical properties such as the mean value of the local intensity distribution or some other statistics such as mean of the maximum or minimum values [21] or local gradient [22], or by splitting the image into subimages and calculating threshold values for the individual sub-images [23]. Some variants of the above two methods can be found in Refs. [17,18].

0 0

Post a comment