Abstract.
A new non-linear filter, called ASLCF(Adaptive-Scale Local Clustering Filter) is proposed, in order to facilitate autonomous image segmentation. The objective is to enhance the edges and simultaneously smooth the uniform regions. This is iteratively accomplished by clustering image blocks of fixed size displaced randomly and changing the current pixel intensities so that the dissimilarity to the corresponding cluster centres is reduced. Uniform intensity and shade regions are smoothed, while edge contrast is increased. The performances are evaluated through the effect of the filter on edges detected by using thresholded zero crossings of a second derivative LoG (Laplacian of Gaussian) operator. Experimental examples of segmented images using the proposed technique are included. The computation time is O(N) and is independent of the scale parameters of the filter.