Accurate and efficient seismic horizon interpretation is important for seismic geomorphology study. By integrating the improved density-based clustering method to generate horizon patches and a heuristic combining strategy to merge horizon patches, we have developed a novel data mining approach to automatically extract globally optimal horizons for detailed geomorphologic interpretation. First, the application of improved density-based clustering method has distinct merits in calculation speed and avoiding the phenomenon of mis-ties. We design a heuristic combining strategy to effectively combine the horizon patches. It is also able to ameliorate the problem of mis-ties that frequently occurs in horizon picking. Second, the proposed algorithm can identify abnormal unit in terms of independent horizon fragments. Furthermore, the introduced method is capable of detecting small-scale seismic geomorphologic features. The applications indicate good real-time performance of our new global interpretation algorithm in automated-tracking speed and quality. Our method can resolve the problem of mis-ties in cases of complex seismic reflection to a certain extent. Besides, not only are a series of channels separately recognized, but also small-scale meandering rivers are clearly mapped. Our algorithm is capable of adding more geologic information and realizing a better showcase of geomorphologic features.