Abstract: The main goal in image enhancement is to accentuate certain image features for subsequent analysis or for image display. In this paper, we describe a new nonlinear filter, called IEC (Iterative Extreme Clustering), intended to facilitate subsequent image segmentation. The objective is to enhance edges while keeping uniform regions as they are. At each iteration, the image is randomly partitioned in a number of fixed size rectangular blocks, which are classified according to their ranges as uniform or nonuniform ones. No changes are made in uniform blocks. Pixels in nonuniform blocks are clustered, with cluster centres obtained by gray level morphological dilation and erosion. Gray levels are then shifted towards corresponding cluster centres, according to a learning parameter. Results and algorithm complexity compare favorably with nonlinear diffusion and related approaches.
Key words: Image enhancement, nonlinear filter, iterative extreme clustering, image segmentation.