Using my movies fingerprint and movies dominant colors scripts, and after I parallelized some of the color-processing routines, I decided to take on a larger project involving a set of movies from the same studio. I’ve watched a lot of Disney animated films, so I took the list of their animation studios films and processed them all individually.
First, I generated all the independent fingerprints with their titles (using the movies fingerprint algorithm with the movies dominant colors upgrades).
The “fingerprint” was generated by re-scaling the movies to a size of
640:370 and extracted
3600 frames from each video file (distributed uniformly in time). From these still frames, the dominant color was calculated from a number of
5 total clusters per frame.
Some examples look like this (click to enlarge):
Then I used the image-grid python package to assemble them together:
I still want to make a slight tweak to the routines so that it allows the use of a clustering algorithm that doesn’t pre-require the number of clusters as an input (such as agglomerative clustering).
As this is a use case of previous coding exercises, please have a look at the original posts (movies fingerprint algorithm, movies dominant colors, and color palette extractor) for the descriptions of how the code works.