The Research Of Image Segmentation Based On Thresholds

Image segmentation is an important content of computer vision and image processing and analysis. Image understanding in computer vision includes target detection, feature extraction and target recognition et cetera, which are all dependent on the quality of image segmentation. Thresholding is a simple but effective tool to separate objects from the background. It is widely used in the image processing of text, quality testing of materials, medical image processing, transportation, military target detection and many other aspects.Otsu method is a frequently used thresholding technology. This paper finds that the clustering results of K-means keep the order of the initial centroids with respect to one-dimensional data set. Based on that, the paper proves that the objective function of Otsu method is equivalent to that of K-means method in multilevel thresholding. They are both based on a same criterion that minimizes the within-class variance. Therefor, the paper proposes a thresholding method based on K-means. However, Otsu method is an exhaustive algorithm of searching the global optimal threshold, while K-means is a local optimal method. If and only if the local optimal threshold is the global threshold, two thresholds found by these two methods are equal. Compared with several thresholding methods, K-means thresholding can get good results in misclassification error and uniformity. Furthermore, the multilevel K-means thresholding is time saving and efficient.One dimensional K-means thresholding only considers the gray levels, the paper also considers the mean of neighbors, and then extends the K-means thresholding to two dimensions. The two dimensional K-means thresholding is robust when there is gaussian noise or salt and pepper noise in images.When there is a lot of noise in an image, two dimensional K-means thresholding can not resist noise effiently, the paper considers gray levels, mean of neighbors and median of neighbors, then extends the K-means thresholding to three dimensions, enhances the ability of resisting noise interference.Because K-means which based on L1 distance is more robust than K-means which based on L2 distance, the paper proposes a K-means thresholding based on L1 distance which can reduce noise interference efficiently.The experiment shows that the result gotten by L2 K-means thresholding is much better than L1 K-means, when there is a lot of noise of high value in an image.