https://bugs.kde.org/show_bug.cgi?id=472416
--- Comment #2 from Michael Miller <michael_mil...@msn.com> --- I looked into this idea when implementing YuNet and SFace. There are 2 concerns with it. First, there is a significantly greater than 0 probability that 2 faces found in the same image could be the same person. Multi-mirror self images, and images where the person and a picture of the person are in the same image. (I have many pictures of my son where there is a picture of him in the background. Both faces were found). When face matching, both faces would be presented to the use as possibilities, and both faces would be correct suggestions. The second issue is the face recognizer as it is currently written has no context of the overall image it came from. The recognizer is given 2 thumbnails and returns a confidence level the thumbnails are the same person. We would have to provide much more information to the recognizer about the source image for this to work. I think introducing this feature could be problematic at best. It sounds like a good idea until you start working through the details. Craig, I hope you'll try the new face detection and recognition features coming in digiKam 8.5.0. The significantly improved accuracy of face matching should help considerably with reducing false-positive suggestions. Cheers, Mike -- You are receiving this mail because: You are watching all bug changes.
