Comparing faces

Alexander (Sasha) Pastukhov

2025-10-09

Tridimensional regression can be used to compare face similarity via R2 based on 3D landmarks.

faces <- 
  list("M010"=Face3D_M010,
      "M101"=Face3D_M101,
      "M244"=Face3D_M244,
      "M92"=Face3D_M92,
      "W070"=Face3D_W070,
      "W097"=Face3D_W097,
      "W182"=Face3D_W182,
      "W243"=Face3D_W243)

face_comparison <-
  dplyr::as_tibble(t(combn(names(faces), 2)), 
                   .name_repair = function(x){c("Face1", "Face2")}) %>%
  group_by(Face1, Face2) %>%
  nest() %>%
  mutate(Fit = purrr::map2(Face1, 
                           Face2, 
                           ~fit_transformation_df(faces[[.x]],
                                                  faces[[.y]],
                                                  transformation ='translation',
                                                  refresh=0)))

face_R2 <-
  face_comparison %>%
  mutate(R2 = purrr::map(Fit, ~R2(.))) %>%
  select(Face1, Face2, R2) %>%
  unnest(cols=c(R2)) %>%
  arrange(desc(R2))

knitr::kable(face_R2, digits = c(0, 0, 3, 3, 3))
Face1 Face2 R2 R2_5.5 R2_94.5
W070 W097 0.988 0.987 0.988
M010 W243 0.984 0.984 0.985
M244 W243 0.984 0.983 0.984
W182 W243 0.980 0.980 0.981
M010 M92 0.979 0.978 0.980
W097 W243 0.978 0.978 0.979
M010 M244 0.978 0.978 0.979
M92 W243 0.975 0.975 0.976
W070 W243 0.975 0.974 0.975
M010 M101 0.975 0.974 0.975
M244 W182 0.975 0.973 0.975
M92 W182 0.973 0.972 0.974
M101 W243 0.973 0.972 0.974
W070 W182 0.972 0.971 0.973
W097 W182 0.971 0.969 0.971
M244 M92 0.969 0.968 0.969
M101 M244 0.967 0.966 0.968
M010 W182 0.967 0.965 0.968
M101 W097 0.965 0.963 0.966
M101 M92 0.962 0.961 0.963
M101 W070 0.962 0.960 0.963
M010 W070 0.962 0.960 0.963
M010 W097 0.961 0.959 0.962
M244 W097 0.961 0.959 0.962
M101 W182 0.955 0.953 0.957
M244 W070 0.955 0.953 0.956
M92 W070 0.948 0.945 0.949
M92 W097 0.946 0.944 0.948 
ggplot(data=face_R2, aes(x=Face1, y=Face2, size=R2, color=R2)) + 
  geom_point()