文档详情

基于时空域的视频去噪算法研究-信号与信息处理专业论文.docx

发布:2019-03-28约6.37万字共67页下载文档
文本预览下载声明
、粤 、粤., 中文摘要 数字视频序列中的噪声不仅会降低视频的主观视觉质嚣,而且会对压缩、 编码、传输和存储等眉续处理产生不利影响,因此数字图像视频去噪技术的研 究对于推动整个数字媒体产业的发展意义重大、影响深远。 本文在对数字图像的嘿声类型、常用的视频去噪算法进行总结的基础上,针 对现有图像质最评价方法不能准确反映图像主观质量的问题,提出了一种基于人 眼视觉特性(Human visual system ,HVS)的客观评价方法。 根据常用数字图像去噪算法以及视频去噪算法,本文提出了…种新的基于 时空域的视频去噪算法。该算法充分利用视频序列空域和时域的相关性,在对视 频序列进行空城滤波的基础上,对其进行时域滤波得到去噪后的视频序列。空 域滤波采用基于即S 的方向性空域滤波算法,通过方向性检测得到空城滤波模 板,并结合人眼视觉特性对该模板进行调整,以得到符合人眼主观视觉的滤波 效果:时域滤波采用基于时域递归滤波的 Kalman 算法来去除噪声影响。 最后对本文所提算法进行了仿真实验,实验结果表明该算法能够有效去除视 频序列中的噪声,提高视频阁像的质量。 关键词:视频去噪; HVS; 质量评价:方向性检测; Kalman 滤波 、/ ‘ 咱 ABSTRACT 、Noise in the digital video s叫uence does not only degrade 伽.e subjective visual quality of video,but also lead adverse effects 阳 the post-processing such ω compressing ,encoding,transporting and storinging and so on . 80 the research on 也e 、 denoising of image s叫uence has si伊ificant meaning and far-reaching influence on promoting the development ofthe whole digi阳1media industry. on the basis of describering 血e types of noises in digital images and commonly 吨 used algorithms for video denoising,也is pap时 proposes a new image quality assessment algorithm based on 伽 human visual syst棚, which can solve the problem that the objective evaluation of image quality is inconsistent with the subjective evaluation. 白lfOUgh the research of the common digital image and video denoising algorithms,a new vidωdenoising algorithm based on temporal叩atial domain has been proposed.ηle algori甘1m makes use of the correlation characterlstics of spatial domain and temporal domain of 由.e image 叫uences. First,spatial filtering is máde for the image sequences,and on 白is basis,the temproral filtering is carried out ωg创 the fmal denoised image s叫uences.ηle 叩atial processing adopts direction spatial filtering algorithm based on human visual system. The spatial filtering template is got by 也e direction detection at frrst,and then the templates are adjusted by the HVS characteristics,ω 也e filtering results according to the human subjective visual 盯e achieve
显示全部
相似文档