Digital image denoising poses a significant challenge in real-time systems due to the presence of high levels of noise and low-resolution images. The inter- and intra-variance between signals during image acquisition contributes to the noisy artifacts in digital images. Various types of noise, such as Gaussian, speckle, impulsive, and combined noise, can be found in images acquired through Synthetic Aperture Radar (SAR) and medical sensors. Traditional denoising techniques like non-linear median filters, Bayesian filters, and wavelet-based shearlet transforms encounter difficulties in effectively analyzing compressed or noisy images, particularly in preserving edge details.To address the issues related to speckle noise, conventional denoising approaches such as Bayesian denoising, non-local filters, wavelet-based shearlet transformations, and autoencoders are commonly utilized. However, these techniques face challenges when dealing with ultrasound images and medical images that contain multiple additive, multiplicative, and Gaussian noise sources. Furthermore, these models struggle to overcome the problem of sparsity in low Signal-to-Noise Ratio (SNR) images.To overcome these challenges, an innovative approach incorporating a hybrid non-linear filter and segmentation-based CNN framework is implemented to enhance denoising performance across various imaging systems. Experimental simulations are conducted on diverse real-time noisy images to assess the efficiency of this denoising approach in comparison to conventional techniques