RAW Noise Reduction Algorithm based on Improved Bilateral Filtering with NLM Method and Implementation on ASIC

Image noise reduction has long been a prominent research area within image processing, and numerous classical filtering algorithms, such as Gaussian filtering and mean filtering, have been proposed by scholars. While these algorithms effectively reduce noise, they also introduce a certain degree of image blurring, compromising the original details. Consequently, researchers have introduced bilateral filtering, NLM (Non-local Means), and other algorithms that aim to preserve the image's detail information while reducing noise. However, these algorithms often have disadvantages such as high computational complexity and slow processing speed, or limited noise reduction capability. To address these limitations, this paper presents an improved noise reduction algorithm called RAWNR (RAW Domain Noise Reduction, RAW means an image obtained directly from the image sensor without any processing). The proposed algorithm combines the respective strengths of bilateral filtering and NLM and adopts a block-based method to enhance noise reduction while accelerating the processing speed. Moreover, the algorithm is implemented using Application Specific Integrated Circuit (ASIC) technology, further optimizing the processing performance and efficiency. The noise reduction image processed by RAWNR exhibit a high PSNR (peak signal-to-noise ratio). Specifically, when applied to a Lena image with a noise intensity of 20, the processed image achieves a PSNR value of 32.35. Experimental results show that the ASIC achieves a frame rate of 60fps when processing images of 5M size (resolution of 2592×1944) at a RAWNR_CLK of 500MHz.