Threshold Segmentation
The threshold segmentation method is divided into global threshold and local threshold. The so-called local threshold segmentation method is to divide the original image into smaller images, and select the appropriate threshold for each sub-image. After threshold value segmentation, the border between adjacent sub-images may produce gray-level non-continuity. Therefore, smoothing technology is required to eliminate this problem. Common Methods of the local threshold method include the gray difference histogram method and the differential histogram method. Although the local threshold segmentation method can improve the segmentation effect, there are several disadvantages:
(1) The size of each sub-image cannot be too small; otherwise, the statistical result is meaningless.
(2) the splitting of each image is arbitrary. If a sub-image falls in the target area or background area, it is divided according to the statistical results, it may produce worse results.
(3) The local threshold method collects statistics on each sub-image, which is slow and difficult to meet real-time requirements.
The global threshold segmentation method is widely used in image processing. It uses a fixed threshold value to segment the entire image. The classic threshold value is used as the processing object based on the grayscale histogram. Based on the different threshold selection methods, the methods can be divided into modal methods, iterative Threshold Selection and other methods. These methods determine the segmentation threshold based on the histogram of the image. In addition, there are inter-Class Variance threshold segmentation, two-dimensional Maximum Entropy segmentation, fuzzy threshold segmentation, symbiotic matrix segmentation, and region growth.
For a relatively simple image, we can assume that the object and the background are in different gray levels, and the image is contaminated by zero mean Gaussian noise, therefore, the gray distribution curve of the image is considered to be composed of two normal distribution functions () and (). The histogram of the image will show two separated peaks, as shown in Figure 5. For such an image, you can select the gray value corresponding to the troughs between the two peaks of the Histogram as the segmentation threshold. This method will inevitably lead to incorrect segmentation, so that some background pixels are judged as objects, and some pixels belonging to objects are also mistaken as backgrounds. It can be proved that when the size of an object is equal to that of the background, the threshold value can be used to minimize the false score probability. In most cases, the histogram of the image is sparse in pixels near the valley, so this method has little impact on image segmentation. This method can be extended to multi-object images with different Gray mean values.
Double-peak Histogram
Iterative Threshold SelectionAlgorithmIs an improvement of the previous method. It first selects an approximate threshold T, divides the image into two parts and calculates the mean sum of the region and the new segmentation threshold t = () /2. Repeat the above steps until and no longer change.