Authentication and identification of physical objects using microstructural features

A system for authentication of a physical object based on microstructural features extracted from patches of a surface of the physical object, the system comprises:

• a backend module including machine learning (ML) models configured to process a plurality of input surfaces of an object by a feature extractor of the ML models using a deep neural network, wherein the plurality of input surfaces varies in dimensions, and the plurality of input surfaces are processed by the deep neural network from a plurality of images of the object captured by a device camera, and
• the backend module is configured to:
• receive by the feature extractor, an input surface from the plurality of input surfaces which is uniformly sized with respect to sizes of the plurality of input surfaces wherein the feature extractor describes the input surface as a one-dimensional feature vector;
• divide the input surface of the object into a plurality of micro-surfaces;
• train the feature extractor on the plurality of micro-surfaces to identify a plurality of microstructural features of the input surface of the object;
• apply the trained feature extractor on a plurality of surfaces corresponding to a plurality of objects, the plurality of surfaces have different dimensions from the plurality of input surfaces;
• store a set of microstructural features of the plurality of surfaces corresponding to the plurality of objects in a database;
• and
• use the trained feature extractor for authenticating a plurality of query objects based on the plurality of microstructural features of the input surface corresponding to each query object, wherein the authentication includes matching the plurality of microstructural features obtained from images of a query object with the set of microstructural features of the plurality of objects stored in the database.

A computer-program product tangibly embodied in a non-transitory machine-readable storage medium, including instructions configured to cause a data processing apparatus to perform operations, for authenticating a physical object based on microstructural features extracted from patches of a surface of the physical object, the operations comprising:

• processing a plurality of input surfaces of an object by a feature extractor using a deep neural network, wherein the plurality of input surfaces varies in dimensions, and the plurality of input surfaces are processed by the deep neural network from a plurality of images of the object captured by a device camera;
• outputting by the feature extractor, an input surface from the plurality of input surfaces which is uniformly sized with respect to sizes of the plurality of input surfaces wherein the feature extractor describes the input surface as a one-dimensional feature vector;
• dividing the input surface of the object into a plurality of micro-surfaces;
• training the feature extractor on the plurality of micro-surfaces to identify a plurality of microstructural features of the input surface of the object;
• applying the trained feature extractor on a plurality of surfaces corresponding to a plurality of objects, the plurality of surfaces have different dimensions from the plurality of input surfaces;
• storing a set of microstructural features of the plurality of surfaces corresponding to the plurality of objects in a database;
• and
• using the trained feature extractor for authenticating a plurality of query objects based on the plurality of microstructural features of the input surface corresponding to each query object, wherein the authentication includes matching the plurality of microstructural features obtained from images of a query object with the set of microstructural features of the plurality of objects stored in the database.