You should not have any difficulty seeing on the figure that the uniform weight leaves the most crosstalk, the nonuniform weights of the linear-estimation method leave less crosstalk, and the nonlinear-estimation method leaves no visible crosstalk. If you cannot see this, then I must blame the method of reproduction of the figures, because the result is clear on the originals, and even clearer on the video screen from which the figure is derived. On the video screen the first iteration is clearly inferior to the result of a few more iterations, but on the printed page these different results are not so easy to distinguish.