Pre-processing plays an essential role in any image processing application. The quality of features is a function of how well the input images are presented to the feature-extracting module or a classifying network. A clean image produces a non-redundant, distinct feature set that assists a classifier system. A variety of noises influence an image due to acquisition systems, interference, and storage. Image denoising, thus being an essential and crucial concern, removes unwanted artefacts from significant details of the image. Autoencoder and Decoder constructed using deep-learned convolutional layers can de-noise images with higher performance metrics. The work in this paper proposes a 7-layer convolutional network (Autoencoder-Decoder) for eliminating noise contents from test images. The Images are from the CIFAR-10 dataset available with the Kaggle data store. The original image is scrambled, encoded using a random key, and subjected to Gaussian noise at the transmitter. It is further decoded with the same key to obtain a noisy image. The noisy images from the input to the convolutional neural network (CNN) and the original images as the targets. The network is trained using 4000 CIFAR-10 images and tested using 1000 images. Experimental analysis showed remarkable results in terms of peak signal-to-noise ratio (PSNR) and structural similarity (SSIM). The average metrics over 10 images showed that the PSNR is about 34.861, and the recovered image retains the original image up to 0.996% (SSIM).