二乘向量机在图像插值技术中的应用研究

图像插值过程实际上是利用已有的像素点来估计与这些像素点相邻的未知像素的颜色。可将一幅图像视为由像素坐标映射而成的二维颜色曲面。这样插值问题就可以视为由已知的像素构成的颜色曲面函数的回归问题。而基于统计学习理论(SLT)的支持向量机(SVM)方法为解决这种高维非线性问题提供了一种新的途经。该文讨论了最小二乘向量机实现数字图像插值的应用方法,并进行了图像放大的试验验证。试验表明:基于RBF核的最小二乘向量机回归技术对数字图像的插值具有很强的适应性,其性能与已有的双立方插值技术相当。

Research on the Application of LS-SVM in Interpolation of the Image

The interpolation of the digital images is virtually a color estimation of the unknown pixels neighboring the available pixels.As images are two dimensional,a color surface can be created from the image that maps the(x,y)coordinates of a pixel to a specific color.In this way,the interpolation of the image is to predict those points lost in the image magnified with the available pixels(sampling points).Due to the high probability that the adjacent pixels will be of a similar color,the color surface should be continuous and can be predicted by the regression theory and method.How to find the optimal color surface function generating the best estimation of the interpolated pixels,the development of the statistical learning theory(SLT)and the support vector machines(SVM)provide us a new approach to develop the accurate interpolation algorithm.The interpolation approach based on the least square support vector machines(LS-SVM)with the RBF kernel is conducted in this paper.Experimentation with the magnification of the different scale images is conducted and the results demonstrate that the standard LS-SVM with RBF kernel has higher adaptability for the image interpolation and near the same estimating accuracy to the traditional cubic algorithm.