Alpha-blending using smoothed mask #72
Conversation
| @@ -76,6 +76,29 @@ cv::Mat convert_rgb_to_yuyv( cv::Mat input ) { | |||
| return yuyv; | |||
| } | |||
|
|
|||
| cv::Mat alpha_blend(cv::Mat srca, cv::Mat srcb, cv::Mat mask) { | |||
There was a problem hiding this comment.
Isn't there something for this already in OpenCV?
There was a problem hiding this comment.
Nothing obvious: it has a constant alpha blend function addWeighted, and masked binary blending via copyTo or add each with a mask parameter.
There is the usual half answered SO question: https://stackoverflow.com/questions/41263010/how-to-blend-images-using-a-mask-in-opencv and an unanswered OpenCV forum one: https://answers.opencv.org/question/28439/blending-two-images-with-masked-roi/
There was a problem hiding this comment.
Searching their "new" forums I stumbled upon:
https://forum.opencv.org/t/opencv-java-how-to-superimpose-three-images-that-have-the-same-size/857/2
This links to this blog post:
https://learnopencv.com/alpha-blending-using-opencv-cpp-python/
There was a problem hiding this comment.
Heh - literally where I got my code from..
|
Initial timing measurements indicate this adds 0.5-1.0 msec to the post-processing compared to |
|
Ah, that's a good idea, thanks. What happens if I cherry-pick this into main, would that mess anything up down the line? |
|
No, this should be fine. Git detects when you merge things where you've had the same patch applied already. |
* Alpha-blending using smoothed mask * Tidy up alpha blending, avoid float math
Using a smoothed segmentation mask to alpha blend between background and foreground (captured video), this is a simple not-quite-light-wrapping (there is no blurred background component) fix to the hard edges we have presently. Might address #67.
For reference: https://www.premiumbeat.com/blog/what-is-light-wrapping-tips-tutorials/ provides a reasonable description of light wrapping.