稀疏表示及其在图像后处理中的应用研究

The problem of image postprocessing includes image denoising, image inpainting, image restoration, image deblocking and image demosaicing, etc. We often need to ehance the distorted-images, resulting from the blocking-artifacts, noise, outliers, and so on, to improve visual quality in many applications including remote sensing, video processing, multimedia communications and biomedical signal processing. The existing image postprocessing methods (DCT, PCA and wavelet analysis) are not able to protect the edges of images. Sparse representation，a new signal processing method developed recently, has been recognized to be promising for image postprocessing because it is a powerful tool for both denoising and deoutliers. However, there are many challenging problems remaining unsolved in image postprocessing by sparse representation. To solve such problems, we systematically investigate sparse representation and its applications to image postprocessing in this project. We aim at developing new efficient image postprocessing methods, which outperform the conventional ones in performance.

图像后处理问题包括图像去噪、图像填充、图像修复、去块效应、去振铃效应等。在影视制作、生物医学信息处理等许多实际应用中，经常需要对降质图像进行以上后处理操作。已有的图像后处理方法(如DCT变换法、主元分析法、小波分析法等)不能很好地处理图像局部化特征，导致其所重构的图像视觉效果不理想。稀疏表示是信号处理领域近年来发展起来的新方法。为克服这一不足，本项目开展基于稀疏表示的图像后处理研究，将稀疏表示理论与技术应用于图像后处理，借助该工具具有从数据中提取隐含内在结构和内在特征能力和降噪、去干扰能力，发展新的图像去噪、图像填充、图像修复等图像后处理技术，克服以主元分析和小波分析等为基础的传统图像后处理方法不便于处理线状奇异性、保护图像边缘等难点问题。

稀疏表示是信号处理领域近年来发展起来的新方法，本项目开展基于稀疏表示的图像后处理研究，将稀疏表示理论与技术应用于图像后处理，借助该工具具有从数据中提取隐含内在结构和内在特征能力和降噪、去干扰能力，发展新的图像去噪、图像填充、图像修复等图像后处理技术，克服以主元分析和小波分析等为基础的传统图像后处理方法不便于处理线状奇异性、保护图像边缘等难点问题。项目主要研究内容如下：.1、用FOCUSS算法等高效的稀疏表示方法替代传统方法，发展新的图像后处理方法。.2、发展有效的“字典矩阵学习算法”，降低对信号稀疏的要求。.3、开展基于非负稀疏表示的图像后处理研究，发展高效的图像后处理方法。