基于多低维模型协同约束的压缩采样图像视频重构

项目名称： 基于多低维模型协同约束的压缩采样图像视频重构

项目编号： No.61471400

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 无线电电子学、电信技术

项目作者： 熊承义

作者单位： 中南民族大学

项目金额： 85万元

中文摘要： 压缩成像技术在军事与民用等诸多领域具有广泛应用前景，而压缩采样图像及视频的高性能重构是其中亟待解决的关键问题，成为当前该领域的研究热点。传统基于稀疏约束的重构方法由于约束模式单一、优化稀疏基受限，因此导致重构质量欠佳。针对该问题，鉴于图像与视频具有更多低维模型并存的特征，拟开展基于多低维模型协同约束的压缩采样图像/视频的重构方法研究，以达到有效提升压缩采样效率和重构性能的目的。主要内容及创新：充分探讨基于块/条带分割的图像及视频的空时域相关及高阶统计相关特性，以低秩表示和流形学习理论为基础，结合分层分割处理与拓扑聚类等技术，探索图像/视频的低秩表示、低维流形表示等更多低维模型；分析图像/视频压缩采样的低秩与低维流形模型的嵌入性能，发展基于稀疏、低秩与低维流形等多低维模型协同约束的压缩采样图像/视频的重构方法与优化实现技术，为压缩采样图像及视频的高效重构提供新思路和有效解决方案。

中文关键词： 图像/视频重构；压缩采样；低秩表示；低维流形；稀疏表示

英文摘要： Compressive imaging has widespread application prospect in many areas in both military and civil fields. High performance recovery for compressive sampled image/video is a key issue needed to address for its successful use, which recently becomes an important research focus hot in this field. Aiming at the problems that the performance of conventional compressive sampling (CS) recovery algorithms only considering sparsity constraint is severly limited because of its single constraint formation and existing difficulty in finding the optimal sparse bases,in this proposal new schemes for enhancing recovery performance of compressive sampling are investigated based on collaborative constraint with multiple low dimensional models when considering more low dimensional models of image/video implicated simultaneously. The main research contents and novelties include: the correlations of image/video in space domain, time domain and high order statistics are probed fully based on patch/stripe division; on the basis of low rank representation and manifold learning theories, more low dimensional models for image and video are explored by comblining layered segmentation process and topological cluster etc.; embedding performence of low rank and low dimensional manifold in CS is analysised in depth, new CS recovery schemes and corresponding optimization technology are developed based on collaborative constraint of multiple low diemnsional models with sparsity, low rank and low dimensional manifold, and finally new thoughts and valid solutions for efficiently reconstructing compressive sensed image/video are provided.

英文关键词： Image/video reconstruction;Compressive sampling;Low rank representation;Low dimensional manifold;Sparse representation