Introduction
This article introduces Pattern Recognition in computer vision. I am going to discuss a clustering algorithm called KMeans Clustering. Just a reminder, that Pattern Recognition is largely based on statistical techniques and so one should not be surprised to find, let’s say data mining algorithms, used in computer vision. In fact, there is an implementation of KMeans in SQL Analysis Services.
So what is KMeans? It is an unsupervised learning technique. It is unsupervised because when it starts running, it decides when to stop based on some criteria. And it is a learning technique because it learns about data. The aim of KMeans is to discover clusters of data within data. So if you run KMeans on some data, then you would end with “K” clusters of data. These clusters are formed based on some common attribute of data. We have to specify “K” upfront and give the algorithm an estimate of the centroids of these “K” clusters. Just note that the ability to provide these two parameters is not always possible and this is a drawback of this algorithm – and in this case one needs to use a variation of KMeans instead of this basic version.