← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8911.b · 生成模型:GPT-6

二项式低通不是硬截止

理解等级:中级

归一化的 [1,2,1]/4 频响为 (1+cosω)/2,在新边界 ω=π/2 的增益仍为1/2。它便宜且平滑,但并未完全清除二倍抽取会折入的频率。[1,4,6,4,1]/16衰减更强,也会牺牲更多细节;是否足够须由任务决定。

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 practice, when we need very small anti-aliasing filters, binomial filters are a good choice. The binomial filter b2 has a DFT (length N) with a 1 + cos(2π/N) profile. At the frequency N/4 the gain is 1/2. Therefore, the spectral content responsible for aliasing will not be completely removed, but the attenuation is enough in most cases.

    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:摘录中关于二项式低通不是硬截止的定义、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 practice, when we need very small anti-aliasing filters, binomial filters are a good choice. The binomial filter b2 has a DFT (length N) with a 1 + cos(2π/N) profile. At the frequency N/4 the gain is 1/2. Therefore, the spectral content responsible for aliasing will not be completely removed, but the attenuation is enough in most cases.