This work is a compagnon project of the GallicaPix PoC. It leverages a dataset of heritage French periodicals built under the GallicaPix scope.
Various face detection tools have been applied to the front page illustrations of the Excelsior newspaper (1910-1943): IBM Watson Visual Recognition, Google Cloud Vision, OpenCV/dnn. See this github. Gender can be infered by some of the tools.
Both the detected faces and the genders have been manually corrected on the page samples used for the quantitative analysis.
The following charts make use of the quantity of faces detected and show the evolution of the Excelsior editorial choices regarding human faces on the front pages, from 1910 to 1920. The categories analysed are :
- men group (from 1 to x men)
- women group (from 1 to x women)
- mixed group (from 1 to x persons)
- couple (one man and one woman)
See some samples here.