基于几何分类的自适应图像插值算法  

A New Adaptive Image Interpolation Method based on Geometric Classification

作　　者：梁云[1,2,3,4] 朱为鹏[2] 李峥[2,5] 

机构地区：[1]华南农业大学信息学院,广东广州510642 [2]中山大学信息科学与技术学院∥数字家庭教育部重点实验室,广东广州510006 [3]深圳市数字生活网络与内容服务重点实验室,广东深圳518057 [4]中山大学深圳研究院,广东深圳518057 [5]广东工业大学计算机学院,广东广州510090

出　　处：《中山大学学报（自然科学版）》2011年第4期11-16,22,共7页Acta Scientiarum Naturalium Universitatis Sunyatseni

基　　金：国家重点基础研究发展计划973资助项目(2006CB30310);广东省自然科学基金资助项目(2009170004203010);华南农业大学校长基金资助项目(2009X029);NSFC-广东联合基金资助项目(U0735001;U0835004;U0935004)

摘　　要：图像插值是放大低分辨图像以适应目标屏幕的有效方法。低分辨率图像边缘特征保持越好,则插值图像的效果越好。根据低分辨图像的边缘分布特征对插值单元几何分类,提出了一种自适应图像插值算法。首先根据高分辨率图像中像素点的相对位置构造矩形插值单元和菱形插值单元,所有未知像素点位于矩形插值单元或菱形插值单元;然后从8个方向,特别是斜对角方向计算插值单元的图像边缘,并将边缘作为割线,根据割线对插值单元进行几何分类,可分为16类;最后根据未知像素点所属的插值单元分类计算未知像素值。实验证明,该算法比现有多种插值算法能够更好的保持图像边缘的尖锐特征。Image interpolation is an effective method to magnify image with low resolution to adapt to the target screen.Better the preservation of image edges brings much better magnifying result.A new adaptive image interpolation approach based on geometric classification of interpolate units,which is formed by edges of image with low resolution,is proposed.Firstly,we construct rectangle interpolate units and rhombic interpolate units by the relationship of pixels in image with high resolution.All pixels of the super-resolution image lie in the two kinds of interpolate units.Then we analysis possible edges of original images especially along the oblique line and treat edges as the secants of interpolate units to classify every interpolate unit.There are total 16 kinds of interpolate units.At last,we calculate unknown pixels based on the classified interpolate units.Experiments show that our new adaptive image interpolation scheme based on geometric classification can preserve edges of interpolated images better than many present methods.

关 键 词：图像插值 几何分类 边缘保持 插值单元 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]