一种改进的放大图像边缘修复算法  

Improved edge inpainting algorithm for image magnification

作　　者：于媛媛[1] 王兆仲[1] 

机构地区：[1]北京航空航天大学图像处理中心,北京100191

出　　处：《计算机工程与应用》2013年第15期167-170,233,共5页Computer Engineering and Applications

基　　金：国家自然科学基金(No.60803071);教育部博士点基金项目(No.200800061067)

摘　　要：为了保持放大后图像清晰且光滑的边缘,提出了一个基于边缘轮廓拟合插值和分水岭变换的边缘修复算法。算法的主要思想是对原始图像进行初始放大,自动确定模糊区域,从而进行修复处理。通过分段三次Hermite插值提取初始放大图像的边缘。对边缘进行膨胀进而自动获取标记图像。对初始放大图像进行距离变换,从而提供了分水岭变换所需的地势图。有了标记图像和地势图就可以利用分水岭变换逐步填充模糊区域像素。实验结果表明:与已有算法相比,该算法降低了对图像梯度的敏感程度,图像边缘更为光滑清晰,并且在运行效率上大大提高。To preserve sharp and smooth edges in magnified images, an algorithm based on contour fitting interpolation and watershed transform is proposed. The main idea of the algorithm is to magnificate the original image, and automatically locate blurry edges, and do the inpainting process. The edges of initially magnified image are extracted by piecewise cubic Hermite inter- polation. A mark image is automatically obtained through edge expansion. A distance transform is applied on the initially magni- fied image to provide the relief image for watershed algorithm. With the help of the mark and relief image, pixels in blurry regions are gradually filled based on the watershed transform. Experiments demonstrate that： compared with some existing algorithms, the method reduces the sensitivity to the image gradient and makes the edges sharper and smoother. Besides, it also has high computational efficiency.

关 键 词：图像修复 图像放大 距离变换 分水岭 像素填充 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]