To address the problems of low accuracy and the extracted coastlines with breakpoints in most existing coastline extraction methods, this paper proposes a regionalized coastline extraction method combined Simple Linear Iteration Clustering (SLIC), Bayes' theorem, and Metropolis-Hastings (M-H) algorithm for high-resolution images. First, the SLIC algorithm is utilized to segment the superpixels for the high-resolution images. Associated with each superpixel, there is a random label, and the label belongs to the land or the ocean. Under the assumptions that all pixels in the superpixel have the same label, that is, the label of the superpixel. On this basis, the Bayes' theorem and the M-H algorithm are combined to define the superpixel labeling optimization function. The optimal results of the function are obtained to achieve coastline extraction after numerous iterations, where the corresponding result of the labeling optimization function is calculated by altering the label of the superpixel in each iteration. Finally, the proposed method is tested on high-resolution images. In the superpixel segmentation, the qualitative results and quantitative evaluation metrics (Boundary Recall and Under segmentation Error) are used to determine the optimal size of superpixel for the image, and the optimal superpixel segmentation is obtained; On the optimal superpixel segmentation, the coastlines are extracted by the Bayes' theorem and the M-H algorithm; From the qualitative results, it can be seen that the proposed method only extract the coastlines, and the extracted coastlines have no breakpoints; According to quantitative evaluation metrics, its average Recall, Overall Accuracy and Kappa coefficient are 99.45%, 99.43% and 0.9882, respectively. Further, the effectiveness and feasibility of the proposed method are demonstrated. To verify the advantages of the proposed method on extracting the coastlines, four compared methods (including Canny operator, object-oriented, BP neural network methods, CNN methods) are tested on high-resolution images. Visually, the Canny operator method not only extracts the coastlines with breakpoints, but also extracts the boundaries of many land targets; the other two compared methods can also extract the boundaries of some land targets except for extracting complete coastlines. Two compared methods, which can extract complete coastlines, are evaluated quantitatively. For the objectoriented method, its average R is 71.24%, the average OA is 84.29%, and the average K is 0.7157; For the BP neural network method, its average R is 91.29%, the average OA is 93.04%, and the average K is 0.8707. For the CNN method, the average R, OA and K are 90.73%, 90.60% and 0.9023, respectively. Further, it can be verified that the proposed method can not only extract complete coastlines, but also has higher accuracy.