A Study In Attraction Through Composite Imagery

I did this one day last year, when I stuck in bed with the flu. I saw a similar process a few years ago done by some university study and I was inspired to give it a go myself. I wanted to see what would happen if I took an average of all the faces of women that I've mentioned on this blog that I've found attractive (as I often think I have an unusual choice in women). Well first off, it's not very extensive as I've only used a handful of images, and it's not exactly very scientific but it is kinda interesting. It's missing a lot of ethnicities (particularly some women of African origins*) and the choice of images arent that great (for instance, I wish had a better pic of Katharine Cullen) you get the idea anyway, and I preferred to have a more honest selection of candid shots than 'shopped to the max glamor pix.

Not going to go into huge detail with the process but basically I merged two images together each time and then merged another image into each result. I did multiple cross mergings of images and resultant images so that a non-weighted average was formed. You can make out most of this process in the video below... (music by me too btw <--- for the copyright police). Faces include: Billie Piper, America Ferrera, Kate Beckinsale, Katerine Cullen, Evangeline Lilly, Selena Gomez, Isabelle Adjani, Danay Garcia and Crown Princess Victoria.



There were three main stages to the process. First, I simply layered each image on top of each other, linearly. Then I took the first merged result and merged it with the last etc. Then I finally merged them all back to back again. (and I think I did a few more mergings <--- shit loads of mergings basically).

Unlike most other studies similar to this process, I choose to alter the shape of the features each time so that they lined up almost perfectly. Most other studies simple lined up hundreds of selected images and let the dominant position of features dictate the position of each feature in the final result. It was nesescery for me to alter the features at times due to the uneven nature of model angles in image sources, plus I had a much smaller sample set of images to work from. I also like the chaotic nature of changes created with each iteration. I also altered the width of the images as I saw fit (<--- scientific eh?)



To finish off I submitted the result to http://myheritage.com to see what happened. Not surprisingly, I got a lot of Latin/Mayan results but interestingly Evangeline Lilly appeared in the results, and Ornella Muti who starred in Flash Gordon, which was the first film I remember seeing, and she was quite nice in it. So that's kinda deeply psychological right? :)

I had planned writing something very scientific, in dept and personal about all this, hence why I delayed posting for so long, but I got sick of seeing this in drafts so this is what you get now. Fantastic!

*But wait! There's more! Since penning this post almost 12 months ago I've gone back and added a few more faces (for the lulz) and deliberately added in the lovely Gabrielle Union as I felt that side of things in particular was missing.



So this is the original picture at the top of this post plus the extra faces of Anja Pärson, Gabrielle Union, MyAnna Buring, and Sienna Guillory. The can see the order or merging in the direction process tree picture thing below.


Click for bigger...


Once again, the image was submitted to MyHeritage.com and here are the results. Lost of Asians appearing this time. Understandable as there appears to be a slight palpebral slant developing in the eyes. Kate Beckinsale appears strongly also. While I don't put much if any faith in MyHeritage.com's recognition algorithms, there does seem to be a similar trend going on with the faces I've chosen.



Finally, the most important question to answer: what do I think of the resultant faces? Yes! Both are fabulous, my dear. I would rate them both highly on my Jackson Pollock scale of facial beauty, the second one in particular acquiring a few extra Jackson Pollocks.