摘要： 为解决监控视频分辨率不足的问题,在视频流体模型的基础上,提出一种图像超分辨率重建方法。视频流体模型记录了视频对象的整体区域,及区域内各像素的时域对应关系,利用流体区域在不同时刻的像素值进行滤波和拼接,达到去噪、扩展分辨率的目的,基于等色线构建视频流体模型,使用视频流体模型实现去噪,以起始帧作为参考图像,并依次在各帧中选择补入流纹,根据补入流纹的相邻流纹计算补入流纹在初始帧的位置,如果所得位置非整数,对参考图像插值拉伸,采用补入流纹中的值代替相关坐标的像素值。实验结果表明,将添加噪声的CIF 格式视频重建到2CIF 格式,该方法的重建结果比最大后验估计与投影方法、梯度投影等方法的峰值信噪比提高1 dB ~4 dB。

关键词: 视频监控, 超分辨重建, 视频流体模型, 视频流量跟踪, 中值滤波, 像素扩展

Abstract: To solve the problem of insufficient resolution monitor videos,this paper presents a video image superresolution reconstruction method based on Video Fluid Model(VFM). A VFM records the region of a video object,as well as the pixels mapping in time domain. Denoising and resolution improving can be achieved by filtering and splicing pixels in VFM regions of different time. VFM is established based on isochromatic line firstly,and video denoising is carried out. The starting frame is selected as a reference image and flow tracking replenished are selected for interpolation frame by frame. The replenished FT is stretched to the first frame,referring to the FT adjacent. Experimental results show that Peak Signal to Noise Ratio(PSNR) gained by this method is about 1 dB to 4 dB higher than other methds,such as Maximum A Posteriori & Projection(MAPP) and Gradient Projection(GP).

Key words: video surveillance, super-resolution reconstruction, Video Fluid Model ( VFM ), video flow tracking, median filtering, pixel expansion

中图分类号: 

• TP391