A new portrait of W.A.Mozart from the
biometrical statistical identification
Braun - December 2005
In 2004 a new candidate for a painted portrait of the adult Wolfgang Amadeus Mozart (1756-1791) was found in a private art collection in Denmark. It could be traced back to another collection that contained several items from the estate of Johann Lorenz Hagenauer (1712-1792), the friend, landlord, and business partner of Leopold Mozart, the father of W.A. Mozart. Independently from the historical analysis, art expertise attributed the painting to the 18th century (see companion publications).
Here, a biometrical statistical comparison of the portrait with the five known authentic adult portraits of W.A. Mozart is reported. The result is that the person shown by the newly discovered portrait has to be considered as the same person as the one shown by the other five portraits, with an error probability of well below one in 300,000. A comparison of the age related facial traits further revealed that the portrait shows W.A. Mozart from a date between 1782 and 1789, probably from around 1785.
1. The five reference portraits
2. Landmark test on non-identity
3. Feature test on non-identity
4. Digital feature test on identity
1) Hanging cheeks (age related trait, missing in M1
Because the seven features are visible at almost all light conditions, their frequency in the general population could be determined by feature counts in public portrait galleries.
From the corpus of 103 adult Caucasian male portrait paintings, which was set up for an earlier study (Braun, 2005 and 2006ab) and which was based on the portraits in the internet archives of the National Gallery of the USA and the Gemäldegalerie in Berlin, all portraits had to be excluded where forehead or lower jaw were either covered (by clothing or beard) or were not painted accurately enough. Further, because traits 1 and 5 can develop in a later period of adult life and Mozart died at 35, all portraits had to be excluded that showed an apparent age of the sitter of clearly above 40. After the exclusions, 30 portraits remained in the corpus. It was then extended by applying the same filters to all portraits in the internet archives of the Musée du Louvre in Paris and the National Gallery in London. Thus, further 13 and 17 pieces, respectively, entered the corpus, raising the total number to 60.
From the earlier study's corresponding corpus of 103 portrait photographs, which had been set up from the random results of Google picture searches, 70 remained after applying the same exclusion rules as described above.
The count of feature frequencies, separately for paintings and photographs, revealed that in both databases trait 1 and 2 appeared in the same face more often than could be expected from their single frequencies. This observation is biologically plausible, because the probability of both traits is likely to increase with the amount of under-skin tissue in the lower jaw. Therefore, the co-occurrence of these two traits in a face had to be considered as a single new trait.
Similarly, the count of feature frequencies, separately for paintings and photographs, also revealed that in both databases trait 5 and 6 appeared in the same face more often than could be expected from their single frequencies. Also this observation is biologically plausible, because the probability of both traits may depend on qualities of under-skin tissue that are similar above and below the eye. Again, the co-occurrence of these two traits in a face had to be considered as a single new trait.
For traits 3 and 4 the frequencies had already been determined in the mentioned earlier study. For the other traits the frequencies were determined by using the new databases. The complete results are as follows:
1 & 2) Hanging cheeks AND horizontal line on chin: 8
% in paintings, 6 % in photographs.
Next, these five features were tested on correlations. Because all tests were negative and because there is also no biological rationale to assume any correlation, the five features have to be considered as stochastically independent. Thus, the frequency of their joint occurrence is computed by multiplication of the single frequencies. The results for the probability that two non-relatives have the five features in common are below one in 370,000 re the painting database, and below one in 5,800,000 re the photo database.
Further, it should be noted that the probability estimates would have become even more decisive, if the many non-digital features that the new face has in common with the other five faces had entered the calculation.
5. Time window for the date of the new face
Trait 5 in the new face is about as strong as in M1,
M2, and M3, but weaker than in M4
and M5. This again places the painting in the window
of 1782-1789. Assuming a smooth development of these two age-related traces
in W.A. Mozart, the painting would reflect his face from a date near 1785.
6. Possibility of post-mortem construction
Addendum March 16, 2008
|Confrontation of the "Hagenauer Mozart" with each of the other five Mozart portraits|