<p>Marine researchers were greatly assisted by underwater image enhancement techniques in revealing important information concealed beneath the surface. Color cast, blurring, and hazy effects are challenging problems that lower the quality of the underwater photos. Numerous investigations were made possible by the emergence of deep learning techniques and the diverse ways in which underwater photographs were obtained. This work proposes an innovative architecture that transforms the deformed underwater image to an enhanced good quality image with no alteration in the texture, information and structure of the image content. The architecture is built on conditional generative adversarial network with channel attention. For training, the suggested model makes use of paired picture datasets that include distorted images and their corresponding reference images. The paired data in UFO-120 and UIEB datasets, which are made accessible to everyone, are used to carry out the experiments. In order to address the difficulties encountered with underwater imagery, input photos from three color spaces—RGB, HSV, and LAB—are considered. The novel embedded attention block in the described architecture reduces the amount of processing power lost on superfluous activation. Mean absolute error loss, content-based loss, perceptual loss, and channel-wise texture loss are combined with conditional adversarial loss to enhance the performance of the architecture. It is evident from the qualitative assessment of the final enhanced image and from the metrics of PSNR, SSIM, and UIQM that the suggested technique performs better than the most recent methodologies.</p>