F.-Y. Hsieh and K.-C. Fan (Taiwan)
Unsupervised classification, watershed transform, data clustering, gravity-space image
In this paper, a novel unsupervised 2-D data classifier using watershed transform is proposed. Conventionally, the task of data classification is performed by applying a clustering algorithm on unsupervised data to form various data clusters first and a certain decision function generation algorithm is then operated to generate the decision functions. Lastly, incoming data is classified by using the decision functions based on the decision theory. For a set of 2-D data without prior knowledge, our proposed method is capable of automatically generating the decision regions, which are experimentally proven to be feasible in classifying incoming data due to the well design of gravity-space image and the morphological behavior of adopted watershed transform. The task of classification can be accomplished simultaneously along with the extraction of decision regions without needing the utilization of decision theory. Experimental results demonstrate the feasibility and validity of the proposed method. A supervised version of our method is also proposed in the paper by adding the mechanism of markers to enhance the classification performance.
Important Links:
Go Back
