摘要
为实现保持图像边缘锐度的快速插值放大,提出了基于图像曲面曲率信息的类双线性插值方法。首先鉴于双线性插值的低通滤波固有特性,引入像素的值变化以构造类双线性插值模型;为获得相邻像素点的方向趋势,将灰度值模拟为地表高程并与二维坐标形成空间曲面;最后以指定方向的曲面的剖面曲率作为图像边缘等结构类型的判别依据,自适应优化两组两像点均值的方向权重从而获得丰富的像素插值先验。合成图像的两倍插值实验显示,峰值信噪比PSNR提高7.8dB,图像框架相似性SSIM改善0.1。本方法计算类型简单且复杂度低,易于定点平台的实时应用。
To satisfy fast image zooming requirement with preserving edge-sharpness,a quasi-bilinear algorithm based on profile curvature of brightness surface is introduced.Firstly,in view of inherent low-pass property of bilinear,an amendatory formula is established via including pixel variation.Then,for obtaining the dominant direction of neighboring pixels,image coordinate and its brightness are treated as surface when gray value being simulated as altitude of earth surface.Lastly,magnitudes of curvatures of specified direction is regarded as adjustment reference of structures type of image,such as edge or texture.Therefore,weights of direction of a pair of two pixels average can be adaptively evaluated according to several thresholds of curvatures.Experiment results of synthetic data demonstrate that,compared with other similar techniques,Power Signal to Noise Ratio（PSNR） gain of image twice-interpolated is 7.8 dB,and improvement of Structural Similarity（SSIM） is 0.1.Especially,arithmetic type of proposed algorithm is simple,and its complexity is relatively low,thus it is convenient to real-time application over fixed platform.
出处
《光电工程》
CAS
CSCD
北大核心
2011年第4期108-114,123,共8页
Opto-Electronic Engineering
基金
国家863高技术研究发展计划资助项目(2007AA802401)
中国科学院西部之光人才培养计划资助项目
关键词
灰度曲面
剖面曲率
曲率驱动
类双线性插值
gray surface
profile curvature
curvature-driven
quasi-bilinear interpolation
