知识卡片

普通读书笔记卡 · 8894.b · 生成模型:GPT-6

差分会放大独立噪声

理解等级:中级

对相邻独立、等方差 σ² 的噪声做差,方差变成 2σ²,因此直接差分可能让真实结构淹没在噪声中。一般形式还包含协方差项,相关噪声不能照搬“翻倍”。本章的应对方式是在求导时引入平滑,以空间分辨率换取更稳定的变化估计。

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

      As shown in figure 18.6, the input is an image corrupted with Gaussian noise. The result of convolving the input noisy image with the kernel [1, –1] results in an output with an increased noise. This is because the noise at each pixel is independent of each other, so when computing the difference between two adjacent pixels we get increased noise (the difference between two

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第18章《图像导数》(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/c018.xhtml) - 结论依据:摘录中关于差分会放大独立噪声的定义、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:As shown in figure 18.6, the input is an image corrupted with Gaussian noise. The result of convolving the input noisy image with the kernel [1, –1] results in an output with an increased noise. This is because the noise at each pixel is independent of each other, so when computing the difference between two adjacent pixels we get increased noise (the difference between two