In bbuchsbaum/pcfacespace: Data and functions for working weith parametric face drawings of Day and Davidenko (2019).

library(grid)
knitr::opts_chunk$set(echo = TRUE, collapse=TRUE,comment = "#>")

Overview

Parametric Face Drawings (PFD; Day and Davidenko, 2019) are schematic, though realistic, parameterized line drawings of faces based on the statistical distribution of human facial feature. Principal components analysis applied to a set of 400 landmarks manually placed on a demographically diverse sample yields a "face space" that can be used for face modeling applications. This R package is a partial port of the Matlab code of Day and Davidenko hosted at the Open Science Framework: https://osf.io/6uds5/

Below we demonstrate some of the basic features of the R package.

Generate and plot a random face from a demographically diverse face space

library(pcfacespace)
face <- pcfacespace::gen_face()
plot(pfd_splines(face))

Morph two random faces

normcoef <- function(x, type="F") {
  x/norm(as.matrix(x), "F")
}

orthogonalize <- function(x,y) {
  y2 = y - ((x %*% y)/(x %*% x))[1,1] * x
}

degtorad <- function(deg) {
  rad <- pi/180 * deg
  rad
}

radtodeg <- function(rad) {
  rad * 180/pi
}

antimorph <- function(x,y, degree=10) {
  x <- normcoef(x)
  y <- normcoef(y)

  rad <- pi/180 * degree
  b <- rad/degtorad(45)
  delta <- x-y
  delta <- normcoef(delta)

  #bvals = seq(0,1,by=.01)
  #out <- sapply(bvals, function(b) {
  #  acos(x %*% normcoef(delta*b + (1-b)*x))
  #})

  normcoef(delta*b + (1-b)*x)

}


distinctiveness <- 7

## generate 60 normally distributed principal component coefficients for face 1
face1 <- normcoef(rnorm(60))
## generate 60 normally distributed principal component coefficients for face 2
face2 <- normcoef(rnorm(60))
face2 <- orthogonalize(face1,face2)
face2 <- normcoef(face2)

## create a "spline" representation of face1, face2, and the average of face1 and face2 (the morph).
f1 <- pfd_splines(gen_face(coef=face1*distinctiveness))
f2 <- pfd_splines(gen_face(coef=face2*distinctiveness))
f3 <- pfd_splines(gen_face(coef=normcoef(face1+face2)*distinctiveness))

grid.newpage()

multiplot(list(f1,f3,f2), nrow=1, ncol=3)

Push two faces away from each other in PC-space using a "face difference"

dval  <- 7
## set number of components (all pcs > 60 will be set to zero)
ncomp <- 60
## generate `ncomp` normally distributed principal component coefficients for face 1
face1 <- normcoef(rnorm(60)) 
## generate `ncomp` normally distributed principal component coefficients for face 2
face2 <- normcoef(rnorm(60))

## make face1 and face2 orthogonal
face2 <- orthogonalize(face1, face2)
## renormalize
face2 <- normcoef(face2)
print(face1 %*% face2)

## move face1 32 degrees away from face2
antiface1 <- antimorph(face1, face2, degree=45)
## move face2 32 degrees away from face1
antiface2 <- antimorph(face2, face1, degree=45)

grid.newpage()
multiplot(list(pfd_splines(gen_face(coef=antiface1*dval, fixed_pcs=1)),
               pfd_splines(gen_face(coef=face1*dval, fixed_pcs=1)),
               pfd_splines(gen_face(coef=face2*dval, fixed_pcs=1)),
               pfd_splines(gen_face(coef=antiface2*dval, fixed_pcs=1))),
               nrow=1, ncol=4)

References

Day, J., & Davidenko, N. (2019). Parametric face drawings: A demographically diverse and customizable face space model. Journal of vision, 19(11), 7-7.

bbuchsbaum/pcfacespace documentation built on Feb. 19, 2021, 12:15 a.m.