The Study Of Image Segmentation Based On Thresholds

Image segmentation is the process and related methods to segment an image into d- ifferent sub-images with different characters and derive some interested objects. It is a key step from image process to image analysis, and it plays an important role in image engineering. And it is also one of the basic and important technique in computer vision. At present, image segmentation has widely used in industry automatization, online pro- duction test, produce process control, document image processing, remote sensing and biomedicine image analysis, bodyguard surveillance and military, sport, agriculture en- gineering and so on.For characteristic of simple and good computational performance and nice stabilization, image segmentation based on thresholds is the simplest and the most useful in all methods, which including Otsu's segmentation method,maximum entropy thresholding method, minimum error thresholding method and their promotion.Otsu's segmentation method is the best useful in all thresholding method. But it only useful to those images which histogram show an obvious double-peak. For this reason, Liu jianzhuang has proposed 2D Otsu's segmentation method. In order to achieve good segmentation results,2D Otsu'segmentation method makes pixel gray to the first dimension and makes neighborhood average gray to the second dimension.But,2D Otsu's segmentation method can not give the good segmentation results to those images which include noise and uneven distribution. In recent years, the average value in the neighborhood has been introduced as the third characteristic. Based on which Jing xiaojun proposed 3D Otsu's segmentation method, and the better result of segmenting has been obtained for the image with low contrast ratio and low SNR. But this method has very high computing time and computational complexity. The computational complexity achieves to O ( L~6 ).To address this issue, algorithm of decomposable 3D Otsu's segmentation method has been proposed in this paper. The foundation of this method is traditional 3D Otsu's algorithm. According to the features of guide rule, 3D Otsu's algorithm is decomposed three 1D Otsu's algorithm. This method not only can give an better result of segmentation, but also can improve computing speed and reduce the computational complexity. The computational complexity of this method is O ( L ).The Otsu's segmentation method is a weighted hard C-mean clustering by proved. But the application of C-mean clustering is need condition. The condition is that the data set must be mission-shaped and equivalent. So, when the condition is not satisfied, the result of segmentation by Otsu's segmentation method will be bad. To address this issue, a weighting 3D Otsu's segmentation method was proposed from the point of view of information divergence in this paper. This method not only can solve the problem of noise, but also can solve the problem of computing time by the idea of decomposition. And the best advantage of this method is weighted in different dimensions. And the weigths can adjust to the most appropriate one. So the weighting 3D Otsu's segmentation method can give the best result of segmentation.