PMM Based Segmentation of Gray-Scale Images

This paper proposes a new mixture-model-based image segmentation method that uses the Pearson system of distribution for representing the model´s different mixture components, thus creating the Pearson mixture model (PMM). Generally, mixture models used for image segmentation assume the component distributions to be Gaussian in nature giving rise to the Gaussian Mixture Models (GMMs). This normality assumption, which in turn reduces the preciseness of the segmentation results, is absent in the PMM. The PMM prepared is subjected to a hard-clustered training. The results obtained on the application of the PMM to image segmentation are compared with the corresponding results obtained by running a GMM technique under similar training. The hard-clustered training approach is adopted as, along with being simpler in application, the computational time taken for calculating the parameters is much less than the corresponding time taken by the soft-clustered training approach.