一种新颖的基于遗传算法的正则化图像插值方法

传统的图像插值方法，包括零阶插值或最近邻插值、双线性插值、立方样条插值等，是先经补零疏化、后经内插滤波实现的。由于这些内插滤波器不能完成理想的低通滤波功能，传统插值图像会增加一定的虚假内容，即导致方块效应、模糊等。另外．由于内插滤波器是确定的，因而这些插值算法就缺乏利用图像本身信息的机制。为了提高插值图像的质量和增强图像的分辨率，首次提出了一种基于遗传算法的正则化图像插值方法。在该遗传算法中，编码采用实值方式，变异采用“引导”方式，适应度评价函数的正则化项采用图像质量评价的一些客观标准。最后，还分析了在遗传算法中怎样直接调整正则化系数，并将实验获得的该正则化系数应用于相同实验条件下的迭代正则化插值算法中，取得了良好效果。实验证明，该方法实用、可行。

A Novel Regularized Image Interpolation Algorithm Based on Genetic Algorithm

Conventional interpolation algorithms of image, such as zero-order or nearest neighbor, bilinear, cubic spline interpolation, can be analyzed in two steps: (i)upsampling by zero filling, and(ii) low-pass filtering. But ideal low-pass filtering of interpolation can not be practically achieved, which results in high-frequency artifacts in the interpolated image. On the other hand, due to the low-pass filters fixed, these algorithms fail to utilize the information of the image itself. In order to improve the quality of the interpolated image and enhance the resolution of it, a novel regularized image interpolation algorithm based on genetic algorithm is proposed in this paper. This genetic algorithm has real-valued coding, the induced mutation operator and the fitness function for evaluation containing the term of some subjective quality measures, so the convergence of the genetic searching in the solution space is very fast. Finally, we analyze how to choose the regularization parameter in the fitness function, and compare the results with that of iterative regularized interpolation algorithm. The experiments demonstrate that the proposed algorithm is practical and applicable.