Computer Vision Foundations · 免费预览
梯度域编辑
将图像存储为其梯度，编辑这些梯度，再用伪逆重新积分，可得到用周围颜色填充孔洞的图像修复，以及在不凭空捏造的前提下增强细节的锐化。
1. Encode, edit, decode
Encode, edit, decode
2. Inpainting by forgetting gradients
Inpainting by forgetting gradients
3. Sharpening by subtraction
Sharpening by subtraction
4. The pattern to keep
The pattern to keep
5. Why zero gradients fill the hole
In gradient-domain inpainting, all gradients inside the masked word region are set to zero before reintegration. Why does the surrounding red color flood into the hole?
6. The arithmetic of unsharp masking
The sharpening filter is twice the identity minus a normalized blur. Why does the DC gain equal exactly one rather than two?
7. Running the sharpening arithmetic
A flat region holds value 10, and the normalized blur returns 8 there because of a nearby edge bleeding in. The sharpening filter computes twice the original minus the blurred copy. Write the output value.