This work proposes a simple and efficient residual convolutional network for Gaussian image denoising. The architecture incorporates hybrid-dilated depthwise convolutions to capture multi-scale spatial context, followed by pointwise fusion and a lightweight refinement layer to maintain spatial consistency. Unlike attention-based or frequency-domain architectures, the network is purely convolutional, interpretable, and computationally efficient. Training is conducted using a residual Charbonnier loss, which stabilizes optimization and preserves edge information across varying noise levels. Extensive experiments on standard benchmarks show that the proposed convolutional neural network achieves competitive PSNR and SSIM while keeping the architecture lightweight and suitable for practical deployment. The main contribution lies in the compact integration of multi-scale receptive-field aggregation and residual learning within a single-stage, lightweight convolutional design.These findings indicate that carefully designed receptive-field aggregation and residual learning are sufficient to deliver strong Gaussian denoising performance without additional architectural complexity.