利用基于内容的网格模型进行图像恢复

Image restoration based on content-adaptive mesh model

将基于内容的自适应三角形网格模型这种图像表达方法应用于图像恢复.在图像恢复过程中,首先提取图像的特征图,并利用Floyd-S teinberg算法和Delaunay三角化算法产生网格,用来表达图像;然后利用正则化方法对网格节点的灰度值进行迭代,从而恢复该节点的灰度值;最后利用已恢复的网格节点对像素点进行Lagrange插值,从而得到恢复后的图像.该方法能对含有噪声的图像进行有效地恢复,试验证明较有约束最小二乘方法性能更好.

A content-adaptive mesh model is applied to the image restoration. During the process, feature-map of the image was extracted at first, followed by the mesh generation for the purpose of image representation using Floyd-Steinberg algorithm and 2-D Delaunay triangulation algorithm. After restoring the nodes iteratively with regularization algorithm, the whole image could be restored by Lagrange interpolation from the values of its nodes. Experimental results show that the new method improves the image quality and it is better than the constrained least squares method.