基于结构相似保真的图像稀疏表示模型

Image Sparse Representation Model based Structural Similarity

过完备图像稀疏表示是一种最新的图像表示模型,采用过完备字典中原子的线性组合形式实现图像的稀疏表示.传统的过完备图像稀疏表示模型采用重建误差的平方和作为保真项.该保真项没有充分考虑到人眼对图像的感知特性,无法度量图像中边缘、轮廓、纹理等局部几何结构的变化.本文基于过完备稀疏表示理论思想,建立了新的稀疏性正则化的图像稀疏表示模型.模型中的正则项约束图像表示系数的稀疏性,保真项采用更符合视觉感知的结构相似性度量.基于正交匹配追踪算法,提出了基于结构相似度的正交匹配追踪算法.实验结果表示,新的模型能够更好地重构图像的结构信息,获得更好的重建视觉效果.

Overcomplete sparse representation is a new image representation model, in which the image is represented by a sparse linear combination of atoms for given redundant dictionary. Traditionally, sparse representation model use mean squared error as a fidelity term. However, the mean squared error fidelity term is not coincident with human visual perception, and can not measure the variations of edge, contour and texture. In terms of sparse represnetations theory, a new sparsity regularized sparse representation model is proposed. The regularization terms constrains the sparsity of image coefficients. The fidelity term is supposed as structural similarity, which is more consistent in visual perception. According to the orthogonal matching pursuit, a new orthogonal orthogonal matching pursuit based structural similarity is proposed. Experimental results show the new model can preserve the structures better and provide better visual quality.