Video Denoising Pipeline Using NLM-NLB GPU Processing

The patent divides the video denoising process into two distinct steps: first applying NLM denoising to obtain intermediate results, then applying NLB denoising to achieve final stable denoising. This segmentation allows each algorithm to focus on specific aspects, improving overall stability while managing execution time through optimized processing stages.

The NLM denoising is performed as a preliminary step before NLB denoising. This preliminary action prepares the input data for the NLB algorithm by removing obvious noise patterns first, which stabilizes the subsequent NLB processing and reduces the computational burden, thereby managing execution duration.

The patent segments the denoising process into two distinct phases with different algorithmic approaches. The NLM phase focuses on accurate patch matching for initial denoising, while the NLB phase handles refinement. This segmentation allows accurate patch matching to be performed only where needed in the first step, rather than throughout the entire process.

The patent applies NLM denoising partially as a first step rather than using it for the complete denoising process. This partial action achieves the benefit of accurate patch matching where most needed while avoiding the excessive computational cost of applying NLM throughout, as the subsequent NLB step handles remaining processing more efficiently.