知识卡片

普通读书笔记卡 · 8952.c · 生成模型:GPT-6

后验结合观测与图像先验

理解等级:中级

去噪把观测写为y=x+g,在噪声模型给出似然p(y|x)、图像模型给出先验p(x)后,寻找使二者乘积最大的x,即MAP估计。目标是干净图像x,不是观测y;省略归一化常数仅适用于它不随待优化x变化的情况。

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

      The goal is to recover the uncorrupted image ℓ[n, m]. The noise g[m, n] is white Gaussian noise with variance . The denoising problem can be formulated as finding the ℓ[n, m] that maximizes the probability of the uncorrupted image, ℓg, called the maximum a posteriori, or MAP, estimate:

    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:摘录中关于后验结合观测与图像先验的定义、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:The goal is to recover the uncorrupted image ℓ[n, m]. The noise g[m, n] is white Gaussian noise with variance . The denoising problem can be formulated as finding the ℓ[n, m] that maximizes the probability of the uncorrupted image, ℓg, called the maximum a posteriori, or MAP, estimate: