Machine learning method and computing device for art authentication

A computer-implemented method for utilizing machine learning to authenticate works of art, the method comprising:

• receiving a test image data file including image data corresponding to an image of a test painting to be authenticated, the test painting thought to be painted by a first artist;
• receiving a plurality of first artist image data files, each first artist image data file including image data corresponding to an image of a successive one of a plurality of first artist paintings, each first artist painting known to be painted by the first artist;
• receiving a plurality of multiple artist image data files, each multiple artist image data file including image data corresponding to an image of a successive one of a plurality of multiple artist paintings, each multiple artist painting known to be painted by a successive one of a plurality of different artists;
• generating a plurality of test painting tiles from the test image data file, each test painting tile including image data for a first number of pixels;
• generating a plurality of groups of first artist painting tiles, each group from a successive one of the first artist image data files, each first artist painting tile including image data for the first number of pixels;
• generating a plurality of groups of multiple artist painting tiles, each group from a successive one of the multiple artist image data files, each multiple artist painting tile including image data for the first number of pixels;
• training a classifier to determine one of a plurality of classes for each first artist painting tile and each multiple artist painting tile, each class corresponding to one of the first artist and the different artists;
• using the trained classifier to determine the class for each test painting tile;
• and
• determining whether the test painting was likely painted by the first artist according to a percentage of the test painting tiles determined to be the class corresponding to the first artist.

A computing device for utilizing machine learning to authenticate works of art, the computing device comprising:

• a processing element programmed or configured to:
• receive a test image data file including image data corresponding to an image of a test painting to be authenticated, the test painting thought to be painted by a first artist;
• receive a plurality of first artist image data files, each first artist image data file including image data corresponding to an image of a successive one of a plurality of first artist paintings, each first artist painting known to be painted by the first artist;
• receive a plurality of multiple artist image data files, each multiple artist image data file including image data corresponding to an image of a successive one of a plurality of multiple artist paintings, each multiple artist painting known to be painted by a successive one of a plurality of different artists;
• generate a plurality of test painting tiles from the test image data file, each test painting tile including image data for a first number of pixels;
• generate a plurality of groups of first artist painting tiles, each group from a successive one of the first artist image data files, each first artist painting tile including image data for the first number of pixels;
• generate a plurality of groups of multiple artist painting tiles, each group from a successive one of the multiple artist image data files, each multiple artist painting tile including image data for the first number of pixels;
• generate a plurality of numerical image content descriptors for each test painting tile, each single artist painting tile, and each multiple artist painting tile, the numerical image content descriptors generated by at least one of UDAT and WND-CHARM;
• calculate a plurality of distances using a distance-based machine learning algorithm, each distance being a vector distance between the numerical image content descriptors of each test painting tile and the numerical image content descriptors of each first artist painting tile from each group of first artist painting tiles and between the numerical image content descriptors of each test painting tile and the numerical image content descriptors of each multiple artist painting tile from each group of multiple artist painting tiles;
• calculate a first artist average distance including an average of the distances between the test painting tiles and the first artist painting tiles;
• calculate a plurality of multiple artist average distances, each multiple artist average distance being an average of the distances between the test painting tiles and the multiple artist painting tiles of each group of the multiple artist painting tiles;
• and
• determine that the test painting was likely painted by the first artist if the first artist average distance is less than each of the multiple artist average distances.

A computer-implemented method for utilizing machine learning to authenticate works of art, the method comprising:

• receiving a test image data file including image data corresponding to an image of a test painting to be authenticated, the test painting thought to be painted by a first artist;
• receiving a plurality of first artist image data files, each first artist image data file including image data corresponding to an image of a successive one of a plurality of first artist paintings, each first artist painting known to be painted by the first artist;
• receiving a plurality of multiple artist image data files, each multiple artist image data file including image data corresponding to an image of a successive one of a plurality of multiple artist paintings, each multiple artist painting known to be painted by a successive one of a plurality of different artists;
• generating a plurality of test painting tiles from the test image data file, each test painting tile including image data for a first number of pixels;
• generating a plurality of groups of first artist painting tiles, each group from a successive one of the first artist image data files, each first artist painting tile including image data for the first number of pixels;
• generating a plurality of groups of multiple artist painting tiles, each group from a successive one of the multiple artist image data files, each multiple artist painting tile including image data for the first number of pixels;
• generating a plurality of numerical image content descriptors for each test painting tile, each single artist painting tile, and each multiple artist painting tile, the numerical image content descriptors generated by at least one of UDAT and WND-CHARM;
• calculating a plurality of distances using a distance-based machine learning algorithm, each distance being a vector distance between the numerical image content descriptors of each test painting tile and the numerical image content descriptors of each first artist painting tile from each group of first artist painting tiles and between the numerical image content descriptors of each test painting tile and the numerical image content descriptors of each multiple artist painting tile from each group of multiple artist painting tiles;
• calculating a first artist average distance including an average of the distances between the test painting tiles and the first artist painting tiles;
• calculating a plurality of multiple artist average distances, each multiple artist average distance being an average of the distances between the test painting tiles and the multiple artist painting tiles of each group of the multiple artist painting tiles;
• and
• determining that the test painting was likely painted by the first artist if the first artist average distance is less than each of the multiple artist average distances.