摘要
论文介绍了一种保持图像边缘特征的插值方法。经典的插值算法常常使插值后图像边缘部分出现模糊或锯齿现象。为了获得图像边缘部分较好的视觉质量,算法对处于图像边缘部分的象素点首先在原图上寻找边缘方向,即利用边缘移动匹配法计算各象素点之间的相似程度来寻找到边缘方向,再使插值沿边缘方向进行。模拟试验证明,无论对机器生成的图像还是对自然图像,这种方法均能很好地保持图像边缘特征,从而使插值后整幅图像的视觉效果大大改善。该算法运算并不复杂,并且对于任何形式的图像几何变形都适用。
In this paper,the authors present an edge preserving interpolation algorithm for digital images.It is well known that classical interpolation methods such as bilinear or bicubic interpolation often suffer from blurring edges or introducing artifacts around edge area.Because edges of an image convey most essential perceptible information to hu-man eyes,it is important to preserve the edge integrity.In order to achieve high performance while keeping modest com-putational complexity,the authors classify pixels into edge class and non-edge class.For non-edge pixels they take a simple interpolation such as bilinear.As for the pixels in the edge area they identify precisely the orientation of edges by using edge-shifted matching method.The edge-shifted matching method estimates the spatial correlations among pix-els in horizontal and vertical directions over a searching region,and compares the results to get the direction of the strongest spatial correlation.This direction indicates the edge orientation.Based on it,the authors can make the interpola-tion along the edge orientation.Therefore they can improve the interpolated image subjective quality,because the most remarkable visible quality improvement comes from the edges that are most sensitive to people's eyes.To evaluate their interpolation scheme ,they implemented it on both synthesized images(such as letters)and natural scenes.Simulation re-sults show that the proposed algorithm can dramatically reduce visual annoying.The other strength of the proposed algo-rithm is that it is suitable for any geometric transformed images.
出处
《计算机工程与应用》
CSCD
北大核心
2003年第6期73-75,134,共4页
Computer Engineering and Applications
