一种图像快速线性插值的实现方案与分析

A Scheme and Its Analysis for Fast Linear Image Interpolation

几何变换是许多图像处理过程的基础.由于线性插值需要较少的运算量,又能在一定程度上保证图像质量,因此,图像几何变换常使用线性插值方法.然而在需要实时处理的场合或运算能力较差的设备上,线性插值的运算速度仍需提高.本文首先总结了数字设备上运算整数化的两条规则.根据这两条规则,在不损失精度的条件下,得到图像双线性插值的整数化方案(三线性插值可同样处理).然后以几何变换中最常使用的仿射变换为例,给出整个图像变换过程的整数化方法及分析.PC机上的实验证明,本文方案相比原始方法执行性能大大提高.此外,本文的方案是高度可并行的.

Geometry transformation is the basis of many image processing procedure. The linear interpolation method need little computation and can maintain image quality to some extent, so it is selected for the image geometry transformation in many cases. But the linear interpolation still need to be speed up when real-time processing is needed or being computed on devices with poor computing capacity. In this paper, two rules are concluded to integefize operands of an expression computed on digital devices. Based on the two rule.s, a scheme is proposed to integefize bilinear image interpolation （ trilinear interpolation can be as the same）, under conditions that the result of which has no loss in precision compared to the result of the original method. Then, the most frequently used transformation, affine transformation,is taken as an example to give a scheme and its analysis to integerize the whole image transformation procedure. The experiment on PC shows that the proposed scheme is much faster than the original method. In addition, the scheme proposed in this paper is highly parallel.