基于投影的单幅图像超分辨率重构算法

利用矩阵的Kronecker积,将现有的向量式成像模型转化为等价的矩阵式成像模型.基于新的矩阵式成像模型,提出了由高分辨图像构成的全空间到超分辨重构问题的解流形的投影算子,利用该算子可以直接得到超分辨重构结果.给出了理论分析.实验结果证实,所提算法能有效地改进超分辨率重构效果,对于标准Lenna图像,其峰值信噪比(PSNR)较双线性插值法和B-样条插值方法提高2.5dB以上,相对于其他算法也有不同程度的改进.

Single Image Super-Resolution Reconstruction Based on Projection

A new single image super-resolution reconstruction algorithm is presented based on projection. Using the Kronecker product of two matrices, the vector based observation model is converted to an equivalent matrix based model. Further more, a projective operator is defined according to the new observation model. Superresolution reconstruction result can be obtained by this operator directly. Meanwhile, this operator can be used to improve the reconstruction impression acquired by other super-resolution algorithms. Theoretic results and the simulations show that the algorithm is valid. For the standard Lenna image,the PSNR value increased about 2.5 dB compared with those of both the Linear and B-Spline interpolation method.