Diffusion-Based Image Sharpening to Reduce Noise

The patent changes the fundamental parameter of the blurring method from Gaussian convolution to diffusion process. This parameter change allows the system to achieve the same sharpening effect through a different mathematical approach that inherently preserves noise characteristics better. The diffusion process uses a different kernel structure and computation method that reduces noise amplification while maintaining edge enhancement capabilities.

The patent substitutes the mechanical convolution operation with a diffusion-based iterative process. Instead of directly applying a Gaussian kernel through convolution, the system uses repeated diffusion steps that propagate pixel values in a manner analogous to heat diffusion. This substitution changes the underlying mechanism from a single-pass linear operation to a multi-pass iterative process that naturally suppresses noise while preserving important image features.

The patent segments the blurring operation into multiple smaller diffusion steps instead of using a single large Gaussian kernel. Each diffusion step operates with a small, fixed-size kernel (typically 3x3), and the desired level of blurring is achieved by repeating this simple operation multiple times. This segmentation transforms one complex computational task into many simple, identical tasks that are easier to execute and optimize.

The patent employs periodic application of the diffusion operation to achieve the desired blurring effect. Instead of performing one complex convolution, the system repeatedly applies the same simple diffusion kernel multiple times. This periodic action allows the computation to be broken into manageable iterations, each contributing incrementally to the final blur level while maintaining computational efficiency at each step.