We study numerically the parallel iteration of Extremal Rules. For four Extremal Rules, conceived for sharpening algorithms for image processing, we measured, on the square lattice with Von Neumann neighborhood and free boundary conditions, the typical transient length, the loss of information and the damage spreading response considering random and smoothening random damage. The same qualitative behavior was found for all the rules, with no noticeable finite sie effect. They have a fast logarithmic convergence towards the fixed points of the parallel update. The linear damage spreading response has no discontinuity at zero damage, for both kinds of damage. Three of these rules produce similar effects. We propose these rules as sharpening algorithms for image processing.
