Abstract: The present disclosure relates to a method for object identification based on a scalar score corresponding to a set of local microstructural features of an object. The plurality of digital images of one or more surfaces corresponding to the object is captured. The set of local microstructural features is extracted from the plurality of digital images. The plurality of objects with a set of local microstructural features is stored in a database. Local microstructural features of a query object are compared with the set of local microstructural features of the plurality of objects stored in the database using a voting function. A score is generated for individual object based on the comparison. A potential match set of the plurality of objects is selected based on the score and a status of the query object is identified based on the potential match set.

Abstract: The present disclosure relates to a method for identifying or authenticating a physical object using computer vision models with machine learning algorithms. The method includes capturing an image of the physical object, focusing on a portion of the physical object in more detail to capture a first analysis image. The method further includes determining a first set of features associated with the first analysis image, generating a first feature-vector for the physical object, comparing the first feature-vector with a second feature-vector, and generating a notification associated with a status for identification or authentication of the physical object based on the comparison. The second feature-vector is generated from a second set of features that are associated with a second analysis image. The first feature-vector is compared with the second feature-vector to determine a match. A notification is generated indicating a status of the machine matching of the physical object.

Abstract: The present disclosure relates to a method for creating a machine vision agnostic protocol for authenticating physical objects using computer vision (CV) and machine learning (ML) algorithms. Multiple images of a physical object are captured in various conditions, including different materials, shapes, cameras, lighting, and alignments. These images form datasets, from which a focus area is auto-selected using ML algorithm based on the object's features. The machine vision agnostic protocol identifies objects using unique similar patterns from the focus areas. To authenticate an image of a query object captured using a camera, multiple images of the query focus area are taken, and the best quality image is selected. Feature vectors are created and compressed into a query physical code using data from the machine vision agnostic protocol. The query physical code is compared with a pre-stored physical code to determine the object's authenticity.

Abstract: The present disclosure relates to a method for generating a fixed-size representation of a surface based on microstructural features. Microstructural features are extracted from surfaces of the physical object using a feature extractor. An input sequence of tokens is determined. A token from the input sequence characterizes the microstructural features of the surface. The input sequence of tokens is augmented with an additional token. The augmented input sequence is iteratively processed using an attention mechanism. Contextual information is exchanged between the tokens of the input sequence of tokens, and the attention mechanism is trained on the processing of the input sequence of tokens to extract the fixed-size representation of the surface of the physical object from a set of arbitrary size representations. An output of the iterative processing of the augmented input sequence is determined as the fixed-size representation of the surface to determine the status of the physical object.

Abstract: The present disclosure relates to a method for authentication of an object based on microstructural features extracted from an input surface of the object. Input surfaces of the object are processed by a feature extractor using a deep neural network from images of the object captured by a device camera. The input surfaces vary in dimensions. The feature extractor outputs an input surface uniformly sized with respect to sizes of the input surfaces. The input surface is divided into micro-surfaces which are used to train the feature extractor to identify microstructural features of the input surface. The feature extractor is applied on surfaces of other objects that have different dimensions from the input surfaces. Microstructural features of the surfaces of the objects are stored in a database. The feature extractor authenticates query objects by matching the microstructural features of a query object with the microstructural features of the other objects.