科研之友
微信
新浪微博
Facebook
分享链接
摘要
The denoising function in digital imaging devices must consider resource consumption and real-time capability in addition to effective noise-removal performance. One commonly used denoising method is pixel similarity weighted frame averaging (PSWFA). In this study, we improve the denoising capability of PSWFA using a pre-filter that consists of a downsampling operator and a small Gaussian filter. Moreover, given that noise in digital imaging devices is signal dependent and is typically modeled as a Poisson-Gaussian distribution, we introduce generalized Anscombe transformation to remove the signal dependency by rendering the noise variance constant. The transformed image can be considered corrupted by an approximately Gaussian noise. To embed our algorithm in hardware, we implement our algorithm on a Spartan-6 FPGA for evaluation. We also compare our algorithm with some existing denoising methods on FPGA. For further evaluation of the denoising ability, the algorithm is compared with some state-of-the-art algorithms that are not implemented on FPGA but have high performance on a personal computer. Experimental results on both simulated noise videos and actually captured low-light noise videos show the effectiveness of our algorithm, particularly in the processing of large-scale noise.
- 出版日期2017-6
- 单位中国人民解放军国防科学技术大学
