Computer Vision Foundations · 免费预览
拉普拉斯算子、锐化与 Retinex
拉普拉斯算子将两个二阶导数合并为一个旋转不变算子，其零交叉标记边缘；其模糊形式模拟早期视觉；对其梯度相关算子作阈值化可将图像分解为反射率与光照。
1. The sum of second derivatives
The sum of second derivatives
2. Edges as zero-crossings
Edges as zero-crossings
3. An early model of the visual system
An early model of the visual system
4. Retinex, seeing paint under light
Retinex, seeing paint under light
5. What the zero-crossing buys
Why did Marr and Hildreth propose Laplacian zero-crossings rather than large Laplacian values as the definition of an edge?
6. Separating paint from light
Retinex takes logarithms of an observed image made of reflectance times illumination. What property of the two factors does the threshold exploit after differentiation?
7. The Laplacian on a ramp
Apply the 1D Laplacian to the ramp 0 1 2 3 at the position of the value 2, meaning compute one times 1 minus two times 2 plus one times 3. Write the result.