一种轮廓平滑图像超分辨率算法  

A contour smoothing algorithm of image super-resolution

作　　者：郑可新[1] 王宁[2] 王海洁[1] 赵旭[1] 

机构地区：[1]哈尔滨工业大学信息网络中心,黑龙江150001 [2]哈尔滨工业大学计算机科学与技术学院,黑龙江150001

出　　处：《微计算机信息》2010年第35期160-162,共3页Control & Automation

摘　　要：由于图像轮廓突变的不连续性,图像的超分辨率会使轮廓出现锯齿或光环,为了减弱轮廓锯齿或光环效应,利用方向滤波来获取图像轮廓各方向的局部信息。在训练库中,将方向信息和高分辨率图像作为先验知识以获取平滑的轮廓。对每一个像素点通过计算此点和邻域小块的相关系数或相似度,运用最近邻查找算法的全局搜索匹配得到相应的高分辨率图像的像素点,产生高分辨率图像的估计,通过优化得到最后的超分辨率图像。Because pixel values of image are not continuity,image super-resolution brings some ＂jaggy＂ and ＂ring＂ artifacts around the edges.For reducing ＂jaggy＂ and ＂ring＂ artifacts around the edges,we propose a super-resolution algorithm based on contour smoothing.To solve contour smooth problem,we use directional filter to get some local direction information of image contours.In the training set,directions information and high resolution image are priori information.For every pixel through compute the related coefficient or similarity of points and patches,use neighbor nearest global search to get relevant high resolution pixel value,generate estimate of high resolution image.Through optimize this image,get super resolution image.

关 键 词：超分辨率 轮廓平滑 方向滤波 全局匹配 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]