Glaucoma, a neuropathic eye disease, is characterized by increased Intraocular Pressure within the retina and ranks as the primary contributor to worldwide blindness after cataracts. Timely diagnosis stands crucial to avoidthe complete blindness. To address challenges related to under-segmentation and over-segmentation in OC and OD segmentation, we present a multi-level attention-augmented U- Net model. This model incorporates different attentions, such as semantic and structural attentions, to refine convolutional features, enabling the model to learn discriminative global-centric features. By applying a conservative smoothing algorithm, we achieve a high ROI extraction accuracy of 98.67%. Moreover, the proposed Multi-level Attention Augmented U-Net architecture achieves a Dice Co - efficient of 0.930 and 0.870 for the partitioning of the OD and OC segmentation correspondingly