Clustering Using Optimized Gaussian Kernel Function

Abstract

Clustering and segmentation algorithms that depend on Gaussian kernel function as a way for constructing affinity matrix, these algorithms like spectral clustering algorithms suffer from the poor estimation of parzen window . The final results depend on this parameter and differ on each time we change it.In this paper we present a new algorithm for estimation using optimization techniques, we construct a vector , each corresponding to i th row in a dissimilarity matrix which is used to construct an affinity matrix using Gaussian kernel function. Our algorithm shows that choosing as the formula 2 = ( , ) 2 ( , ) 2 is the opti-2 ( , ) 2 ( , ) 2 mum estimation, and we introduce more than one approach to calculate global value for from this vector. The affinity matrix which is produced using our algorithm is very informative and contains addition information like the number of clusters .