0

I have a project to implement a Snapchat Lenses-like face recognition and distortion algorithm. So far I've tried Android's default face detection API, the play-services-vision face detection API, and OpenCV/JavaCV, but they seem to only detect the location of faces and features and not describe the exact shape of the faces.

Is there anything I missed from these libraries that will allow me to do total face recognition that does describe the exact shape of the faces?

P.S. Should I ask this in Superuser instead?

Gensoukyou1337
  • 1,507
  • 1
  • 13
  • 31

1 Answers1

0

That is a problem with a solution still in progress in the area of Digital Image Processing (very good solutions available, but not the best one defined yet).

I can think of one simple solution, that might work. Once you have the features of the face (such as eyes, lips, etc), you can retrieve the color of the skin surrounding it (using a pre-defined window, considering the size of the eyes/lips block). Then, you use the retrieved color to feed a border/edge recognition algorithm.

All of this can be easily done with OpenCV, but I cannot assure the method's accuracy, since I have not tried this myself.

Also, maybe Canny's method could be useful for your application.

But I strongly suggest you to search for contemporary papers on this subject, as methods for face recognition are getting better and better, but it can lead to very complex and efficiency-expensive alternatives.

Take a look at the link below, and search for any paper that leads a method in doing what you want:

Use keywords such as "face border recognition", "border recognition", "face edge detection", or similar. It's highly probable someone has already done this and then you won't have to reinvent the wheel!

Vinícius Queiroz
  • 849
  • 1
  • 7
  • 19