← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8958 · 生成模型:GPT-6

单调映射调整响应分布

理解等级:中级

子带直方图匹配通过逐点单调非线性改变响应大小,把初始高斯噪声的较平缓分布推向参考纹理的尖峰分布。很多系数会变小,某些离散实现也可能变零,但连续拉普拉斯密度本身不包含零点概率质量,不能把尖峰直接解释为多数系数严格等于零。

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

      Figure 28.8 shows how one of the subbands gets transformed by the histogram matching procedure. The histogram of the subband of the reference texture has the Laplacian shape typical of natural images. Most of the values of the output of the oriented filter are zero, and only near the triangle boundaries the values are non-zero. In the first iteration, the texture synthesis process starts with

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第28章《纹理》(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/c028.xhtml) - 结论依据:摘录中关于单调映射调整响应分布的定义、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:Figure 28.8 shows how one of the subbands gets transformed by the histogram matching procedure. The histogram of the subband of the reference texture has the Laplacian shape typical of natural images. Most of the values of the output of the oriented filter are zero, and only near the triangle boundaries the values are non-zero. In the first iteration, the texture synthesis process starts with