Add scenes recognition for automatic indexing

Using Visual Genome (https://visualgenome.org/) data, one could recognize scenes in the pictures and extract keywords for automatic indexing of pictures.

This seems so easy: https://visualgenome.org/api/v0/api_beginners_tutorial.html

The good thing with Visual Genome is that they are using Wordnet Synsets, where translations in Italian, French and other languages exist (http://compling.hss.ntu.edu.sg/omw/) and there is good support in Python, which is not the case with the names used by Google OpenImage (https://storage.googleapis.com/openimages/web/index.html).

The concepts detected in the pictures can be added to a automatic_keywords tag in the album.ini file. Then the scanner will use the words from the manually indexed keywords and automatic_keywords to create the search index. And perhaps in the future, one could stop creating manual keyword entries...