← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8898 · 生成模型:GPT-6

锐化来自原图减模糊

理解等级:中级

若平滑核 g 的系数和为 1,可令锐化核 h=2δ−g,使输出为原图加上“原图减模糊图”的细节残差,直流增益仍为 1。书中示例用归一化的 3×3 二项式核。该运算增强已有频率,不会恢复输入中完全缺失的真实细节。

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

      A simple way to design a sharpening filter is to de-emphasize the blurry components of an image. By the linearity of the convolution operator, we're allowed to add and subtract kernels to make a new kernel that would give us the same filtered image as if we had added and subtracted the filtered outputs of each of the component kernels. For this example, we start

    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:摘录中关于锐化来自原图减模糊的定义、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:A simple way to design a sharpening filter is to de-emphasize the blurry components of an image. By the linearity of the convolution operator, we're allowed to add and subtract kernels to make a new kernel that would give us the same filtered image as if we had added and subtracted the filtered outputs of each of the component kernels. For this example, we start