IRON is a low level operator dedicated to the estimation of single and multiple local orientations in images. Previous works have shown that IRON is more accurate and more selective than Gabor and Steerable filters, for textures corrupted with Gaussian noise. In this paper, we propose two new features. The first one is dedicated to the estimation of orientation in images damaged by impulsive noise. The second one applies when images are corrupted with an amplitude modulation, such as an inhomogeneous lighting.