← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8954 · 生成模型:GPT-6

滤波边缘乘积需要归一

理解等级:高级

把各滤波器各位置的响应边缘密度相乘,可以作为鼓励自然响应的图像先验势函数。但滤波响应往往相互依赖,尤其冗余滤波器组不能把该乘积直接当作已归一的联合密度。若全是零和带通滤波器,常数方向还未受约束,需要灰度范围或低频项等补充。

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

      Let's consider a filter bank of K filters with the convolution kernels: hk[n, m] with k = 0, …, K – 1. The filter bank could be composed of several Gaussian derivatives with different orientations and orders, or Gabor filters, or steerable filters from the steerable pyramid, or any other collection of band-pass filters. If we assume that the output of each filter is independent

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第27章《统计图像模型》(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/c027b.xhtml) - 结论依据:摘录中关于滤波边缘乘积需要归一的定义、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:Let's consider a filter bank of K filters with the convolution kernels: hk[n, m] with k = 0, …, K – 1. The filter bank could be composed of several Gaussian derivatives with different orientations and orders, or Gabor filters, or steerable filters from the steerable pyramid, or any other collection of band-pass filters. If we assume that the output of each filter is independent