<p>Visibility of an image plays an important role in all computer vision based applications. Haze is an environmental condition that adversely affects image visibility. Poor visibility due to haze affects the functionality of different computer vision based applications. Thus, development of an efficient dehazing system is essential to overcome such kind of problems. In this paper, an end-to-end Morphological Fusion Attention Network (MFA-Net) is proposed for image dehazing. The proposed morphological network is developed using parallel paths of an alternate sequence of dilation and erosion operations that form closing or opening layers and employ appropriate structuring elements. The structuring elements are learnt through back-propagation algorithm from input image. Attention mechanisms (channel attention and spatial attention) are applied on the morphological feature maps to improve the visibility of the dehazed output image. The model is evaluated on various benchmark hazed datasets and it has been observed that the proposed hybrid approach outperforms other state-of-the-art techniques both quantitatively and qualitatively with significant less number of parameters. Ablation studies have been performed on performance gain to verify the key design of the proposed image dehazing model.</p>