Feasability Check for a project

solomonB's icon

Hi,
My last project worked out wonderfully, thanks to pmpd and this message board. I am now working on another project that needs to be completed before December 31st. I was wondering if you guys could point me in the right direction and give me some ideas on how feasible this idea is.

I need to make composite faces out of people detected in a video. Only one face would show up at a time in the video. I could use cv.jit.faces to find the face. I could then use alpha masks eyes, nose, mouth, cheeks, etc which would be scaled and applied within the cv.jit.faces box. I would need to scale, position, and compile the various alpha masks so that I have one alphamask which will fit over the face in the original video stream. I would then need to take a snap shot and scale it to my desired final image size so that it could be averaged into my original portrait. I want to use alpha masks for each facial feature so that people of various shapes of face will not cause strange blurs in the final composite image.

Does this sound like the right way to do what I am trying to do? Is this possible in my time frame and how would I go about positioning,scaling,and layering the facial feature masks so that they can be applied to my source images?

Thanks for any help!

cap10subtext's icon

Could be overly ambitious in the time you have. But then again, others might disagree. Might want to try limiting it to a single facial feature or part of the face, then see if it's doable.

I'm curious as to how you'd go about applying the alpha masks.

solomonB's icon

I need to figure out a way to take the part of my input matrix which falls into the box created by cv.jit.faces and scale it to my final composite size. After that I can apply alpha masks to get the facial features I am looking for, but I am stuck at the first step of scaling the initial matrix.

jazmatajz's icon

click twice on cv.jit.faces.draw.to see the structure for drawing the box.
hope it helps

solomonB's icon

Oh wonderful. I didn't realize cv.jit.faces.draw was an abstraction. This makes a world of difference.