知识卡片

普通读书笔记卡 · 8913.d · 生成模型:GPT-6

插值三角核必须对称

理解等级:中级

整数倍率 k 的一维线性插值核可写 h[n]=max(1−|n|/k,0)。绝对值使其关于原样本位置对称,邻点之间的权重按距离变化。书中式21.30漏写绝对值,会让负索引侧权重大于1,与图示及线性插值要求不符,不能原样实现。

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

      Bilinear interpolation: As we discussed before, bilinear interpolation is done in two stages: first, linearly interpolate each sample with the two nearest neighbors across rows, and then do the same for the columns. This can also be written as a convolution with the kernel (if k is even):

    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:摘录中关于插值三角核必须对称的定义、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:Bilinear interpolation: As we discussed before, bilinear interpolation is done in two stages: first, linearly interpolate each sample with the two nearest neighbors across rows, and then do the same for the columns. This can also be written as a convolution with the kernel (if k is even):