知识卡片
离散高斯必须重新核对增益
直接采样连续高斯并截断后,离散系数和一般不等于连续积分。实际均值滤波应按保留系数的和重新归一化;常用约3σ范围在截断处幅度约为中心的1.1%,这不是说二维总截断质量也恰为1%。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 模糊滤波器
原书出处 · 1 处
- · 查看原书依据
We have removed the normalization constant as the sum of the discrete Gaussian will be different from the integral of the continuous function. So here we prefer to define the form in which the value at the origin is 1. In practice, we should normalize the discrete Gaussian by the sum of its values to make sure that the DC gain is 1.
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第17章《模糊滤波器》(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/c017b.xhtml)
- 结论依据:摘录中关于离散高斯必须重新核对增益的定义、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:We have removed the normalization constant as the sum of the discrete Gaussian will be different from the integral of the continuous function. So here we prefer to define the form in which the value at the origin is 1. In practice, we should normalize the discrete Gaussian by the sum of its values to make sure that the DC gain is 1.