← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8914 · 生成模型:GPT-6

上下采样核的增益不同

理解等级:中级

二倍插零后的一维线性插值核是 [1⁄2,1,1⁄2],其和为2;二倍下采样的平滑核 [1,2,1]/4 的和为1。前者补偿插零造成的采样密度下降,二维可分离插值核和为4。外形相似不代表可直接互换归一化,否则会改变亮度。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      In the special case where k = 2, the 1D bilinear interpolation kernel reduces to the binomial filter h[n] = [1/2, 1, 1/2]. And in 2D, for k = 2, the kernel is [1/2, 1, 1/2] ◦ [1/2, 1, 1/2]T. The binomial filter is not the best interpolation filter, but it is very efficient. When quality is important, there are better interpolation filters.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第21章《图像下采样与上采样》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c021.xhtml) - 结论依据:摘录中关于上下采样核的增益不同的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:In the special case where k = 2, the 1D bilinear interpolation kernel reduces to the binomial filter h[n] = [1/2, 1, 1/2]. And in 2D, for k = 2, the kernel is [1/2, 1, 1/2] ◦ [1/2, 1, 1/2]T. The binomial filter is not the best interpolation filter, but it is very efficient. When quality is important, there are better interpolation filters.