GTMS: A Simultaneous Mode Seeking and Clustering

A mode seeking algorithm not only can automatically find mode of density of a given data but also can be used for data clustering. However, finding the mode of all data points produce redundant computations. In this paper, a simultaneous mode seeking and clustering which is called Generalized Transport Mean Shift (GTMS) algorithm was proposed. An idea of transportation was used for remedying the problem. For each iteration, the "transporter-trailer" characteristic was assigned to data points. The data points that tend to be shifted through the same trajectory, by considering shift direction, will be transported by the same transporter. Then they will be excluded from the computation in the next iteration. The transporters were, then, computed for finding a mode of density. The proposed algorithm was evaluated on clustering and image segmentation problems. The experimental results show that GTMS algorithm outperforms existing algorithms in both accuracy and time-consuming. It reduces redundancy computation by excluding data points more than 90% of image segmentation data after the second iteration only.