Thermal and visual image registration (translation)

问题

I am working on image registration between infrared (LWIR) and visual images using FLIR dataset.

RGB LWIR

Scaling was a global transformation problem, which I resolved lucky. However, the translation varies with different image pairs (visual and IR) due to different FoV of visual sensors.

I have tried area based (intensity) image registration which works well with half of the images only. I have also tried few feature based image registration techniques, but results were worse due to heterogeneity of visual sensors data (images).

Currently, I am trying to use Canny or Sobel edge detector to map the edges into matching points to perform control point registration, but I am still far to get good results.

Is there any feasible solution to resolve this translation problem?

Any suggestion/help will be highly appreciated.

Thanks

来源：https://stackoverflow.com/questions/54258930/thermal-and-visual-image-registration-translation