基于统计分析的视频干扰辨识应用研究-计算机系统结构专业论文.docx

万方数据 万方数据 摘 要 随着光电、信息处理和多媒体通信技术的飞速发展和逐步完善，视频信息已广泛应用 于人类的日常生产和生活，视频行业的大数据时代也已悄然而至。然而，由于受到信息获 取设备的安装环境和数量、信息处理算法、传输带宽和储存成本等因素的影响，视频质量 的监控、诊断以及恢复需求成为了目前的当务之急。干扰辨识研究是视频质量监控、诊断 和恢复工作的重要组成部分，能够改变传统的人工质量监控方式和被动的监控系统维护方 式，实现智能监控和智能诊断，并能够为质量恢复工作提供先验知识。条纹干扰、压缩块 效应干扰和时间域上的闪烁干扰是影响视频图像质量的三大常见干扰类型，本文从统计分 析的角度分别对这三类干扰类型进行了分析并给出了有效的无参考辨识方案。 条纹干扰辨识主要利用了条纹干扰在时间域上具有单一方向的运动特性和傅立叶变 换的平移特性。一段时间内，根据傅立叶幅度频谱中峰值点位置的统计特性即可辨识条纹 干扰。 块效应干扰辨识主要利用了块效应干扰边界效应的伪周期性。在已知块大小和块边界 方向的前提下，通过统计固定频率点的幅度和相位频谱特性，提取了有效的干扰辨识方案。 闪烁干扰辨识主要分析了闪烁干扰差分信号的统计特性，通过实验发现其差分信号服 从拉普拉斯分布，在已知分布的前提下采用小概率事件的思想和迭代拟合方案有效地去除 了影响闪烁干扰程度估计的运动目标差分信号。同时，在设计闪烁干扰辨识指标时融入了 视觉掩盖理论。 从充分的实验数据可以观察到提出的辨识方案具有一定的应用价值。 关键词：视频质量评价；条纹干扰；块效应干扰；闪烁干扰；统计分析 I Abstract With the rapid development and gradual improvement of optical, information processing and multimedia communication technology, video information has been widely used in our daily production and life. The era of big data in video industry has quietly arrived. However, due to the impact of the installation environment and quantity of information acquisition equipment, the information processing algorithm, transmission bandwidth and storage costs, etc al. the needs to monitoring, diagnosis and restoration of video quality has become a current priority. Interference identification research is an important part of video quality monitoring, diagnosis and restoration, which is able to change the traditional way of artificial quality monitoring and the passive way of surveillance systems maintaining, to accomplish intelligent monitoring and diagnosis, to provide a priori knowledge for restoration. Stripe interference, blocking interference and flickering interference in temporal domain is three common interference types affecting the quality of video images. From the perspective of the statistical analysis, these three types of interference have been analyzed, and an effective no-reference identification solution has been presented. The movement in one direction of stripe interference in temporal dom