An energy-efficient and high-PSNR image denoising pre-processor based on optimized non-local means algorithm for real-time FHD video applications

Image is widely used in many fields as an important information carrier. In embedded devices, denoise as an important part of image processing will obviously affect the result. The progress of sensor resolution and light-sensitive capability has higher requirements for denoise speed and quality. Most of the embedded cameras can only perform noise reduction based on simple local denoising with unsatisfactory results due to resource constraints, while the non-local denoising with better results has high computational complexity and is difficult to achieve real-time processing in embedded applications. In this paper, an energy-efficient and high-PSNR image denoising pre-processor based on optimized NLM algorithm is proposed with following features: i) optimized parameters in the weight function and the utilized method noise to improve the denoising performance; ii) a customized systolic array with full pipelining to achieve high processing throughput; iii) a data reuse strategy to reduce the frequent data movement for high energy-efficiency. The achieved results of 22.8 mJ/frame and 765μJ/frame in 110-nm ASIC and Xillinx Zynq FPGA shows that the proposed design is a competitive energy-efficient solution for many mobile FHD video applications.