Deep learning performs as a computational tool with various potential utilities in ophthalmology. Retinal infections of the eye need to analyze small retinal vessels, microaneurysms, and exudates in the diagnosis of retinal diseases. Due to the appearance of various noises in the fundus images, the retinal vasculature is too complicated to be analyzed for retinal conditions. In this work, we have focused on the field of advanced deep learning in which plethora of architecture is available with the increase in dimension and flexibility of the retinal fundus images. Removal of noise is an essential part to better visibility of noisy fundus and thus a deep learning method for degraded retinal fundus image restoration scheme has been suggested in this investigation. A deep convolutional denoising auto-encoder method based on total variational multi-norm loss function minimization with batch normalization approach has been introduced for restoration of the fundus. The proposed scheme is utilized to restore the perceptible structural details of fundus as well as to decrease the noise level. Moreover, the speed of the network for target noisy images is faster compared to that of other models after fine-tuning of the network with the dropout mechanism. The retinal image databases such as DRIVE, STARE, and DIARETDB1 have been adopted to assess image denoising effects. Our approach to increase the visibility of fundus images by noise reduction through a deep training method has significantly delivered better performance without losing image details along with having fast convergence rate.