项目名称: 基于结构化集学习的视频稀疏编码理论与技术
项目编号: No.61271218
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 无线电电子学、电信技术
项目作者: 熊红凯
作者单位: 上海交通大学
项目金额: 76万元
中文摘要: 本申请项目为适应日益推广的移动信息平台、多媒体传感网对高清视频信源的交互内容服务,探索在非稳态系统约束中的视频信号逼近性能,确定其非对称的分解(映射编码器的信源构造和描述)、重构(映射解码器的信息合成)框架,分析随机信源(如视频)在信息空间内的高维相关性描述,通过结构化模型的统计学习进行信号集合的稀疏编码,由非确定的输入信号状态构造一定概率区间的最优输出,以利于视频在广义网络化应用中的基本描述和鲁棒传输。提出基于高维特征空间描述和预测的稀疏编码方法,建立基于多维尺度字典学习的低码率视频编码理论与算法,实现基于结构化集学习的编码预测模型,进行基于矩阵填充的正则化视频重建。研究广义多维信息融合的视频通信,使总体的失真特性可以实现最优逼近的模型化描述。本项目从实际应用中抽象出科学问题,涉及信息论、信号处理的基础,预期的理论和技术成果可有助于高性能视频编码和多视信源的发展。
中文关键词: 稀疏编码;子空间分析;结构化预测;字典学习;正则性
英文摘要: To match the increasingly development of interactive high-definition video dissemination over promising mobile platform and multimedia sensor network, we aim to investigate advanced video approximation with stochastic constraints. With the asymmetric analysis-synthesis structure, it is dedicated to fulfilling sparse coding by structural set model and high-dimension correlation. To benefit the fundamental description and robust transmission over ubiquitous network-oriented application, the optimal solution of video coding is made possible in a high probability from uncertain conditions. To be concrete, it is achived by sparse representation and correlated prediction within high-dimension feature space, the multi-dimension and -scale dictionary learning, the structured set prediction model, and the regularized matrix completion. The overall rate-distortion characteristics of video coding is expected to be close to the theoretical bound of nonlinear approximation by the fusion of multi-dimension side-information. It is founded on the combinatorial area of information theory and signal processing, and the achievements could favor the ongoing high-performance video coding standard and multiview video.
英文关键词: Sparse Coding;Subspace Analysis;Structured Prediction;Dictionary Learning;Regularization