Segmentation of retinal anatomical features such as optic nerve head (ONH) and optic cup, the brightest area in the center of ONH which is devoid of neural elements, is a prerequisite for computer-aided diagnosis and follow-up of glaucoma. The ONH segmentation methods, which imposed shape and intensity constraints, are unable to identify ONH and optic cup boundaries at the same time. On the other hand, recent efficient supervised learning-based methods, which provide a unified system, require combination of many informative features as their inputs, as well as ground truth for the training phase. This paper uses a saliency map including color, intensity and orientation contrasts as the input of a winner-take-all neural network, and color visual features as the input of a self-organizing map neural network to segment ONH and optic cup, simultaneously. Our method is evaluated on a database of 205 ocular fundus images provided by local eye hospitals and publicly available image databases RIMONE and DIARETDB0 comprising 60 non-glaucomtous and 145 glaucomatous images. The ground truth is provided by two expert ophthalmologists. The method attained an average overlapping error of 9.6% and 25.1% for ONH and cup segmentation, respectively. Cup-to-disc area ratio (CDR) is computed for glaucoma assessment. The mean and standard deviation of the CDR differences between our method and the ground truth in all images are 0.11 and 0.09, respectively.