基于梯度的快速图像插值算法

Fast image interpolation algorithm based on gradient

为解决传统插值算法效果不佳而新发展的边缘方向插值算法计算量大的问题,提出了一种基于梯度检测边缘的快速图像插值算法。该算法通过在原图像每一个3×3邻域内计算二阶梯度,检测边缘是否存在:若存在边缘,则计算一阶梯度判断边缘强弱并自适应得到权值因子,然后进行带权的线性插值;若不存在边缘,则在此邻域内进行双线性插值。实验结果表明,此算法不仅使插值后图像具备整体清晰性和边缘平滑性,有效保持了图像的主观视觉质量,而且相比现有同等效果的插值算法,运算时间至少能降低20%,尤其适用于嵌入式设备图像放大的应用。

Traditional image interpolation algorithms have poor performance and most of the recently developed edge- directed interpolation algorithms are of high computation. To solve those problems, a fast image interpolation algorithm based on gradient distribution detection was proposed. The basic idea was to detect the existence of edges in each 3×3 neighbourhood of the original low-resolution image by calculating its two-step gradients. If an edge existed, adaptive weighted factor up to edge strength was obtained by calculating its one-step gradient, and then the weighted linear interpolation was implemented. Otherwise, bilinear interpolation was carried through in that neighbourhood. The experiments show that the proposed algorithm is able to not only keep the subjective visual quality of the interpolated high-resolution image with whole clarity and edge smoothness, but also reduce the operation time by at least 20% compared with the equivalent algorithms now available. It contributes to an image enlargement application in embedded systems.