It is recommended to take it with the video:
2.canny edge detection (canny edge detection)_bilibili_bilibili
1. Common convolution kernels
Summary of commonly used image convolution kernel types_Gaussian convolution kernel-CSDN Blog
1 Gaussian convolution kernel
Two-dimensional normal (Gaussian) distribution
It is linearly separable, can be smoothed, and can remove random noise.
Pixels will be affected by surrounding pixels, so using a Gaussian convolution kernel can remove noise to a certain extent.
2 Sobel convolution kernel
2. Edge detection
1 Gaussian kernel convolution denoising:
2 Edge judgment:
The edge has the greatest rate of change. The derivation is the largest, but there is no way to derivation for pixels because it is a discrete value. Therefore, the difference quotient is used instead. The pixel distance is the same, so the pixel difference is also the same.
This is the principle of sobel.
3 Non-maximum suppression:
Linear interpolation method simulates g1 and g3. See whether point c is the maximum value.
4 Grayscale threshold setting:
Remove some extreme values that may not be edges: dual threshold setting
Edge connection: See that there is a reserved extreme value in its eight fields, connecting them both. The data structure uses a stack.
3. Code learning
I haven’t learned yet, but I have learned to write.