Underwater images often lack clarity due to the attenuation and dispersion characteristics of light. This causes degraded quality and loss of information in underwater images. In this paper, a novel architecture for improving the quality of underwater images by serially combining two deep learning networks, namely Color Correction Network (CCN) and Perceptual Enhancement Network (PEN) is proposed. The former network is used for restoring colors in the image whereas the latter one is used for perception enhancement of the image. A combined loss function of Mean Squared Error (MSE) and Mean Gradient Error (MGE), is used in training the model. This combined loss function helps in obtaining images that maintains overall image quality and preserves structural sharpness. The proposed architecture helps in obtaining visually superior underwater images which is demonstrated using both subjective and objective analyses.