For successful image segmentation, it is vital to discover optimal discriminating features that contrast one region from the other, or contrast salient edges from the background. We propose a new method for image segmentation based on three discriminating features: average gradient magnitude, uniformity of gradient magnitude and uniformity of gradient direction across a range of scales. The problem of threshold selection has been avoided by partitioning the feature space into edge and background clusters. Experimental results show that the combination of these three features possesses significant discriminating power to separate edges from the background.