Mean shift clustering uses sliding-window to find dense areas in the data points. It is a centroid-based algorithm. The goal of the algorithm is to locate the centre points of each group/class, which works by updating candidates for centre points to be the mean of the points within the sliding-window. These candidate windows are then filtered in a post-processing stage to eliminate near-duplicates, forming the final set of centre points and their corresponding groups. The assumption is that the population is dense at the centre of each cluster.
Mean-shift algorithm starts by selecting a random point in the data space. Then, check the data in a fixed radius around the point, to identify the direction where the density increases most. The point continues to shift in the data space in order to move in the direction where the density increases most. This continues till you reach a point where the density decreases on every side. This is the centre of the group. This process is done with many sliding windows until all points lie within a window. When multiple windows overlap the merge and the maximum is preserved. The data points are then clustered according to the sliding window in which they reside.
This is different from the K-means algorithm because we do not select the number of clusters. The mean shift algorithm identifies this for us. That adds a lot of value. Ofcourse it leaves the room for configuration (hence room for doubt) because we have to select the radius of the sliding window. The radius in turn impacts the count of clusters identified.
This limitation of the algorithm is that it identifies the cluster based on the density of points and if an entire cluster is not dense enough, it is pushed into another dense cluster. Thus, it may miss some clusters. But the algorithm offers a very good computational performance.