基于最小二阶导数的低复杂度视频背景建模算法  

Video background modeling algorithm with low complexity based on the minimum second derivative

作　　者：张安伦 滕国伟[1] 赵海武[1] 李国平[1] 王国中[1] ZHANG An-lun;TENG Guo-wei;ZHAO Hai-wu;LI Guo-ping;WANG Guo-zhong(School of Computer and Information Engineering,Shanghai University,Shanghai 200072,China)

机构地区：[1]上海大学通信与信息工程学院

出　　处：《光电子．激光》2018年第8期858-864,共7页Journal of Optoelectronics·Laser

基　　金：国家高技术研究发展计划(863计划)(2015AA015903);上海市自然基金项目(14ZR1415200)资助项目

摘　　要：为了训练出适应于视频压缩域的高质量背景模型,该文根据像素点在时域上的分布特征,提出一种基于最小二阶导数的低复杂度视频背景建模算法。首先,根据函数的二阶导数性质来判断其波动特性;然后,通过二次差分拟合像素点在时域上的二阶导数,得到各个像素点的稳定性;最后,根据设定阈值分离出各个位置的像素点在训练周期内最平稳的值,将其作为相应位置的背景模型值。实验结果显示,与AVS2相比,BD-rate节省了9.83%,BD-PSNR提升了0.37dB。与AVS2-S的背景建模算法相比,本算法有效改善了前景污染问题,降低了算法复杂度。In order to train a high-quality background model adapted to the video compression domain, this paper proposes a video background modeling algorithm with low complexity based on the minimum sec- ond derivative according to the distribution of pixels in the time domain. Firstly,estimate the wave char- acteristics of the function according to its second derivative. After that,get the stability of every pixel by using the second-order difference to fit the second derivate of pixels in the time domain. Finally,extract the steadiest value of every pixel during the training period in the basis of threshold value,then take it as the corresponding background model value. The experimental results indicate that compared with AVS 2, BD-rate is saved by 9.83%, and BD-PSNR is improved by 0. 37 dB. Compared with the background modeling algorithm of AVS2-S,this algorithm not only effectively improves the problem of foreground pollution,but also reduces the algorithm complexity.

关 键 词：视频编码 AVS2 AVS2-S 背景建模 二次差分 

分 类 号：TN919.81[电子电信—通信与信息系统]