Computer Vision Foundations · 免费预览

导数的离散化

连续导数变成一个小卷积核,后向差分平移半个采样,而中心差分以分辨率换取对齐。

1. From limits to taps

From limits to taps

2. The half-sample shift

The half-sample shift

3. The Fourier verdict

The Fourier verdict

4. Two dimensions and history

Two dimensions and history

5. Where the backward difference files its answer

The kernel d0=[1,−1]d_0 = [1, -1] compares sample nn with sample n−1n-1. At which location does its output naturally belong, and what does that imply on a step edge?

6. What the central difference gives up

The central difference d1=[1,0,−1]/2d_1 = [1,0,-1]/2 has response jsin⁡(2πu/N)j\sin(2\pi u/N). What does this response do at the highest image frequency?

7. Running the backward difference

The signal reads 5 then 7 then 2 across three samples. Apply d0=[1,−1]d_0 = [1, -1] and write the first valid output value, the one that compares the second sample against the first.

进入交互式练习

免费注册 · 即时批改 · 自动安排间隔复习。

免费开始这门课