微光场景下的时空域EMCCD视频降噪算法

Spatio-Temporal Domain EMCCD Video Noise Reduction Algorithm in Low Light Scene

针对EMCCD(Electron-Multiplying Charge-Coupled Device)在微光条件下捕获的视频噪声大(峰值信噪比低于25 dB)的问题,提出一种降噪效果好且计算复杂度低的改进时空联合降噪算法EMVDE(EMCCD Video Denoising,用于EMCCD的视频降噪算法)。EMVDE在时域上执行改进卡尔曼滤波与像素相似度加权帧平均结合的算法进行运动估计和消除帧间噪声,空域上执行双边滤波、直方图均衡以及去斜条纹算法进行细节增强。在数据集上的实验结果表明,与原始大噪声视频帧(光照度0.002 lx)相比,降噪后视频帧的峰值信噪比平均提升了9.18 dB,结构相似度平均提升了0.43,视觉效果得到了明显提升;同时,EMVDE算法基于FPGA硬件实现仅占用30%的LUT和36%的I/O资源,工作频率高达22MHz,能够满足以25 fps的速率处理视频序列的需求。

Aiming at the problem of high noise in the video(peak signal-to-noise ratio less than 25 dB)captured by EMCCD under low light condition,an improved spatio-temporal joint denoising algorithm(EMCCD Video Denoising,EMVDE)was proposed,which has good denoising effect and low computational complexity.EMVDE implemented an algorithm combining improved Kalman filtering with pixel similarity weighted frame averaging in time domain to estimate motion and eliminate interframe noise.In spatial domain,it performed bilateral filtering,histogram equalization and diagonal stripe remove algorithm for detail enhancement.The experimental results on the data set show that compared with the original high-noise video frame(Illuminance 0.002 lx),the peak signal-to-noise ratio of the denoised video frame is improved by 9.18 dB on average,the structural similarity is increased by 0.43 on average,and the visual effect is improved obviously.At the same time,based on FPGA hardware,the EMVDE algorithm realizes only 30%of LUT and 36%of I/O resources,and its working frequency is as high as 22 MHz,which can meet the requirement of processing video sequence at 25 fps rate.