基于压缩感知理论的图像采样、编码和重建研究

With the significant increase of image data, image resolution, and the applications of taking pictures in mobile devices, the existing image coding and decoding technology is currently stuck in bottleneck and facing new challenges. Therefore, exploring the breakthrough of the existing structure in theory and establishing a new generation of image codec framework have become an urgent and core problem. This project will study a new and efficient image coding and decoding technology based on the theory of compressive sensing (CS), and will carry out the work from three aspects, i.e. sampling, coding and reconstruction. The concrete content is as follows. For the convenience of hardware implementation and cost savings, a simple and effective 0-1 binary sparse CS random measurement matrix is designed. Exploiting the inherent characteristics of natural images, a new sparse random measurement matrix which is able to generate image CS measurements with high correlations is presented, and the corresponding high efficiency coding scheme is developed. Based on the high spatial correlations in natural images, an adaptive spatially directional predictive coding scheme for CS measurements is proposed, with the aim of improving the efficiency of image coding. To take full advantage of natural image properties, a more compact image representation with structural sparse model is built, and a soft CS reconstruction objective functional considering CS quantization process is established. To achieve high-quality image CS reconstruction, an efficient and robust iterative optimization algorithm is accordingly proposed. This project will make theoretical innovations and technological breakthroughs, and will further greatly promote the development of the new generation of image coding and decoding technology based on the CS theory.

随着图像数据量的显著增大、图像分辨率的成倍提高、移动设备拍照等一些应用的日渐普及，现有的图像编解码技术在发展中遇到了瓶颈和新的挑战。因此，探索从理论上突破现有架构，建立新一代的图像编解码框架已经成为一个亟需解决的核心问题。本项目将研究基于压缩感知理论的新型高效图像编解码技术，分别从采样、编码与重建三个层面开展。具体内容包括：设计简单有效的基于0-1二值稀疏压缩感知随机测量矩阵，便于硬件实现和节省成本；根据自然图像固有特性，设计能够获取高相关性测量值的稀疏随机测量矩阵，研究高效编码方案;针对自然图像中存在的高度空间相关性，研究预测方向自适应的压缩感知测量值编码，提高图像编码效率；建立具有更加紧致表达的图像结构稀疏模型，构造考虑量化过程的图像压缩感知软重建目标函数；设计高效鲁棒的优化问题迭代求解算法。本项目可取得理论创新与技术突破，为基于压缩感知理论的新一代的图像编解码技术的研发起到积极推动。

随着图像数据量的显著增大、图像分辨率的成倍提高、移动设备拍照等一些应用的日渐普及，现有的图像编解码技术在发展中遇到了瓶颈和新的挑战。因此，探索从理论上突破现有架构，建立新一代的图像编解码框架已经成为一个亟需解决的核心问题。本项目研究基于压缩感知理论的新型高效图像编解码技术，分别从采样、编码与重建三个层面开展。主要研究内容包括：根据自然图像固有特性，设计能够获取高相关性测量值的稀疏随机测量矩阵；建立具有更加紧致表达的图像视频结构稀疏模型，构造考虑量化过程的图像压缩感知软重建目标函数；设计高效鲁棒的优化问题迭代求解算法，获得了主流最好的效果。另外，在重建处理方面，提出了新型的基于聚类和协同表示的超分辨率算法以及基于限制非凸低秩模型的去块效应算法；在编码压缩方面提出了自适应运动矢量精度预测算法等。本项目共发表高水平学术论文9篇，都是本领域的国际顶级期刊和国际顶级会议，包括2篇T-IP,3篇T-CSVT,1篇T-MM和3篇DCC。