Cluster Analysis by Binary Morphology
IEEE Transactions on Pattern Analysis and Machine Intelligence
Self-organizing maps
| Hi-index | 0.00 |
In this paper, we present a new data classification algorithm in an unsupervised context, which is based on both Kohonen maps and mathematical morphology. The first part of the proposed algorithm consists to a projection of the distribution of multidimensional data observations onto a Kohonen map which is represented by the underlying probability density function (pdf). Under the assumption that each modal region of this density function has a correspondance with a one and only one cluster in the distribution, the second part of the algorithm consists in partitioning the Kohonen map into connected modal regions by making concepts of morphological watershed transformation suitable for their detection. The classification process is then based on the so detected modal regions. As an application of the proposed algorithm, the sample of observations is constituted by image pixels with 3 color components in the RGB color space. The purpose is to present a new approach for unsupervised color image classification without using any thresholding procedure.