Suppose I morph two faces and save the results to 11 jpg files. Naively I would expect that picture number 05 will contain 50% of face oe and 50% of face two, right?
Yes, it should. The pixel values are weighted 50% for each imge in the center.
But it is not always true. Sometimes I can clearly recognize person 2 in 50/50 picture and even more in 60/40. So, my question, do you know how to define the border (I mean 50/50 border) between two arbitrary faces?
I can think of two reasons reason for this. It could be a subjective effect - perhaps because the features of the one face are more distinct. It could also be becuase the contrast in one of the pictures is higher. You can adjust contrast and brightness in the Color Mapper.
And one more, could you give a reference for the algorithm you use?
The algorithm is one I developed myself. Actually I think of it as two different algorithms: One for connecting the points in a triangular grid, and one to do the actual morphing.
I can't say much about the first. It is something I made 5 years ago, and I hardly even remember how it works. It makes the grid so that the triangles are as close to equilateral as possible to avoid long sharp triangles.
The morphing is quite simple: The points are just moved on a straight line from the position in image 1 to the position in image 2, while the texture inside is linarely transformed from that in image 1 to that in image 2. The technique is very similar to that used in 3D texture mapping.
> software developer <