从图像降质模型出发，研究运用最大后验概率(MAP)估计法实现图像超分辨率重建。简单介绍了MAP方法的发展现状，并分析了该算法中存在的缺陷，即目标函数的吉布斯(Gibbs)项对于重建图像的噪声抑制力不均衡。针对该缺陷采用原始低分辨率图像插值后图像的梯度场对MAP目标函数的Gibbs项系数进行修正，使该系数对各像素根据相应梯度值自适应的调整，在一定程度上均衡了目标函数对于不同梯度值区域的约束力。采用共轭梯度法对改进前后MAP算法分别求解并进行了仿真。结果显示相比传统MAP算法，改进的MAP算法得到的超分辨率图像,既很好地恢复了细节，又很好地抑制了重建过程中引入的噪声，总体像质有了明显提高，同时在迭代求解过程中也表现出很好的收敛性与稳定性。

The theory of image degraded model and maximum a posteriori probability (MAP) are introduced in brief. Then the defect of the MAP, the constraints of the Gibbs term (the second term of the objective function) to pixels with different gradients unbalanced are analysed. Based on this defect, a modified MAP algorithm for reconstruction is presented. The gradient matrix gotten from the interpolated image of the low-resolution image to modify the Gibbs term is used, so that the constraints are balanced to some extent. Then the updated MAP objective function is minimized by conjugate gradient method and the modified algorithm is simulated. The results show that, compared with the original MAP algorithm, the modified MAP algorithm can keep details well, and control noise (generated in reconstruction process) greatly in the reconstruction, and the image quality is improved obviously. Meanwhile, the modified MAP algorithm is steady and convergent in solving the problem.