Stone Inscriptions are ancient handwritten scripts that comprise heritage information and are engraved on the stone walls of ancient temples and other structures. In digital stone paleography, the digitization of stone in the form of images and enhancing the image clarity for further text recognition is essential. These images have a lot of background noise due to stone texture, especially some noise patterns that look similar to parts of the characters. The existing denoising techniques are not efficient for these types of images because the disparity between background and foreground is very small, hence it is necessary to formulate a new approach. This study proposes DDNS, a hybrid denoising technique that combines the features of Discrete wavelet transform, Denoising autoencoder, Non-local means and Sauvola thresholding. The autoencoder is trained with numerous samples of different stone inscription images with different ambient lighting conditions, background colours, textures, and foreground characters with different writing depths. The efficacy of this approach is evaluated using the metrics peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM), and the experimental results demonstrate the effectiveness of our system in eliminating noise patterns and improving the recognition process.