知识卡片

普通读书笔记卡 · 8910 · 生成模型:GPT-6

下采样折叠多个频谱

理解等级:高级

对可整除的 N×M 图像沿两轴抽取 k 倍,采用书中未归一化正向DFT时,输出频谱是 k² 个按 N/k、M/k 周期偏移的原谱副本之和,再乘1/k²。它不是简单裁剪频谱;高频副本落入保留区域就会与原有低频混合。

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

      Now we are ready to write the DFT of . This is the DFT of the product of two images, equation (21.9), so we can apply the dual of the circular convolution theorem for finite length discrete signals (where both signals are extended periodically):

    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:摘录中关于下采样折叠多个频谱的定义、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:Now we are ready to write the DFT of . This is the DFT of the product of two images, equation (21.9), so we can apply the dual of the circular convolution theorem for finite length discrete signals (where both signals are extended periodically):