System for determining biometric representation data
A machine learning network is trained to use biometric input data to generate representation data used to assert identity. The network may be trained to consider particular features of interest. For example, the biometric input data may comprise an image of a user's hand. Particular features may comprise one or more of fine print ridges on the hand, principal lines and creases, wrinkles, veins, and so forth. In one implementation intermediate data produced by a first portion of the network is trained to be representative of the particular feature(s) of interest. The resulting representation data, such as produced by a second portion of the network, may thus be based on these features of interest. In another implementation the image is processed to determine feature data, and the image and feature data are provided as input to the network. A hybrid network may also be used.
1 . A system comprising:
one or more memory devices storing first computer-executable instructions; and
one or more hardware processors to execute the first computer-executable instructions to:
determine a first set of training data comprising first training input image data, associated first training feature data, and associated label data, wherein:
the associated first training feature data comprises at least:
first information associated with a first type of feature of an object depicted in the first training input image data, and
second information associated with a second type of feature of the object depicted in the first training input image data;
provide the first training input image data as input to a first machine learning network (“first network”);
determine, based on the first training input image data and using a first portion of the first network, first intermediate data, wherein the first intermediate data corresponds to respective types of feature data;
determine, based on the first intermediate data and the associated first training feature data, a first loss value using a first loss function;
determine, based on the first training input image data and using a second portion of the first network, first representation data;
determine, based on the first representation data and the associated label data, a second loss value using a second loss function;
determine a first weight associated with the first loss value;
determine a second weight associated with the second loss value;
determine a third loss value based on the first loss value, the first weight, the second loss value, and the second weight;
determine trained model data based at least in part on the third loss value, wherein the trained model data is associated with operation of the first portion and the second portion of the first network;
acquire second input image data;
determine second representation data, based on using the first network and the trained model data to process the second input image data; and
use the second representation data to perform one or more of:
determine enrollment user data, or
determine asserted identification data.
2 . The system of claim 1 , the one or more hardware processors to further execute the first computer-executable instructions to:
normalize the first loss value and the second loss value to a common range; and
wherein the first weight is less than the second weight.
3 . The system of claim 1 , wherein the associated first training feature data and the first intermediate data are representative of one or more of:
skin friction ridges of a hand,
principal lines of the hand,
creases of the hand,
wrinkles of skin of the hand,
veins of the hand, or
shape of the hand.
4 . The system of claim 1 , the one or more hardware processors to further execute the first computer-executable instructions to:
process the first training input image data using a feature extraction module to determine the associated first training feature data.
5 . The system of claim 1 , wherein:
the enrollment user data comprises previously stored representation data that is associated with identification data; and
the asserted identification data is determined based on a comparison of the second representation data and the previously stored representation data.
6 . The system of claim 1 , the one or more hardware processors to further execute the first computer-executable instructions to:
provide the associated first training feature data as input to the first network;
wherein the first portion of the first network determines the first intermediate data further based on the associated first training feature data; and
wherein the second portion of the first network determines the first representation data further based on the associated first training feature data.
7 . The system of claim 1 , wherein the first training input image data and the associated first training feature data are associated with a first modality; and
the first set of training data further comprising:
second training input image data associated with a second modality; and
second training feature data associated with the second modality; and
the one or more hardware processors to further execute the first computer-executable instructions to:
determine second intermediate data based on the second training input image data and using a third portion of the first network;
determine, based on the second intermediate data and the second training feature data, a fourth loss value using a third loss function;
determine, based on the second training input image data and using a fourth portion of the first network, third representation data;
determine, based on the third representation data and the associated label data, a fifth loss value using a fourth loss function;
determine a third weight associated with the fourth loss value;
determine a fourth weight associated with the fifth loss value; and
determine the third loss value further based on the fourth loss value, the third weight, the fifth loss value, and the fourth weight.
8 . A system comprising:
one or more memory devices storing first computer-executable instructions; and
one or more hardware processors to execute the first computer-executable instructions to:
determine trained model data, wherein:
the trained model data is associated with first training feature data corresponding to two or more types of features of an object depicted in first training input image data,
the trained model data is associated with operation of multiple portions of a first machine learning network (“first network”), and
the trained model data is based on:
a first loss value that is associated with first intermediate data that corresponds to the two or more types of features of the object,
a second loss value that is associated with first representation data that is based on the first training input image data, and
a third loss value that is based at least on the first loss value and the second loss value;
receive first input image data;
determine second representation data, based on using the first network and the trained model data to process the first input image data;
determine enrolled user data comprising previously stored representation data that is associated with identification data; and
determine asserted identification data based on a comparison of the second representation data and the previously stored representation data.
9 . The system of claim 8 , the one or more hardware processors to further execute the first computer-executable instructions to:
determine a first set of training data comprising the first training input image data, the first training feature data, and associated label data;
provide the first training input image data as input to the first network;
determine, based on the first training input image data and using a first portion of the first network, the first intermediate data;
determine, based on the first intermediate data and the first training feature data, the first loss value using a first loss function;
determine, based on the first training input image data and using a second portion of the first network, the first representation data;
determine, based on the first representation data and the associated label data, the second loss value using a second loss function;
determine a first weight associated with the first loss value;
determine a second weight associated with the second loss value;
determine the third loss value based on the first loss value, the first weight, the second loss value, and the second weight; and
determine the trained model data based at least in part on the third loss value, wherein the trained model data is associated with operation of the first portion and the second portion of the first network.
10 . The system of claim 8 , the one or more hardware processors to further execute the first computer-executable instructions to:
determine a first set of training data comprising the first training input image data, the first training feature data, and associated label data;
provide the first training input image data and the first training feature data as input to the first network;
determine, based on the first training input image data and the first training feature data and using the first network, the first representation data;
determine, based on the the first representation data and the associated label data, the second loss value using a loss function; and
determine the trained model data based at least in part on the second loss value, wherein the trained model data is associated with operation of the first network.
11 . The system of claim 10 , the one or more hardware processors to further execute the first computer-executable instructions to:
determine, using the first input image data as input to a first feature extraction module, first feature data; and
determine the second representation data, further based on using the first network and the trained model data to process the first feature data.
12 . The system of claim 8 , the one or more hardware processors to further execute the first computer-executable instructions to:
determine a first set of training data comprising the first training input image data, the first training feature data, and associated label data;
provide the first training input image data and the first training feature data as input to the first network;
determine, based on the first training input image data and the first training feature data and using a first portion of the first network, the first intermediate data;
determine, based on the first intermediate data and the first training feature data, the first loss value using a first loss function;
determine, based on the first training input image data and the first training feature data and using a second portion of the first network, the first representation data;
determine, based on the first representation data and the associated label data, the second loss value using a second loss function;
determine a first weight associated with the first loss value;
determine a second weight associated with the second loss value;
determine the third loss value based on the first loss value, the first weight, the second loss value, and the second weight; and
determine the trained model data based at least in part on the third loss value, wherein the trained model data is associated with operation of the first portion and the second portion of the first network.
13 . The system of claim 12 , the one or more hardware processors to further execute the first computer-executable instructions to:
determine, using the first input image data as input to a first feature extraction module, first feature data; and
determine the second representation data, further based on using the first network and the trained model data to process the first feature data.
14 . The system of claim 8 , wherein the first input image data is representative of one or more of:
skin friction ridges of a hand,
principal lines of the hand,
creases of the hand,
wrinkles of skin of the hand,
veins of the hand, or
shape of the hand.
15 . The system of claim 8 , wherein the first input image data comprises a first image acquired using a first modality and a second image acquired using a second modality.
16 . A method comprising:
determining a first set of training data comprising first training input image data, associated first training feature data, and associated label data, wherein:
the associated first training feature data comprises at least:
first information associated with a first type of feature of an object depicted in the first training input image data, and
second information associated with a second type of feature of the object depicted in the first training input image data;
providing the first training input image data as input to a first machine learning network (“first network”);
determining, based on the first training input image data and using a first portion of the first network, first intermediate data, wherein the first intermediate data corresponds to respective types of feature data;
determining, based on the first intermediate data and the associated first training feature data, a first loss value using a first loss function;
determining, based on the first training input image data and using a second portion of the first network, first representation data;
determining, based on the first representation data and the associated label data, a second loss value using a second loss function;
determining a first weight associated with the first loss value;
determining a second weight associated with the second loss value;
determining a third loss value based on the first loss value, the first weight, the second loss value, and the second weight;
determining trained model data based at least in part on the third loss value, wherein the trained model data is associated with operation of the first portion and the second portion of the first network;
acquiring second input image data; and
determining second representation data, based on using the first network and the trained model data to process the second input image data.
17 . The method of claim 16 , wherein the associated first training feature data and the first intermediate data are representative of one or more of:
skin friction ridges of a hand,
principal lines of the hand,
creases of the hand,
wrinkles of skin of the hand, or
veins of the hand.
18 . The method of claim 16 , further comprising:
determining enrolled user data comprising previously stored representation data that is associated with identification data; and
determining asserted identification data based on a comparison of the second representation data and the previously stored representation data.
19 . The method of claim 16 , further comprising:
providing the associated first training feature data as input to the first network;
wherein the determining the first intermediate data by the first portion of the first network is further based on the associated first training feature data; and
wherein the determining the first representation data by the second portion of the first network is further based on the associated first training feature data.
20 . The method of claim 16 , wherein the first training input image data and the associated first training feature data are associated with a first modality; and
the first set of training data further comprising:
second training input image data associated with a second modality; and
second training feature data associated with the second modality;
determining, based on the second training input image data and using a third portion of the first network, second intermediate data;
determining, based on the second intermediate data and the second training feature data, a fourth loss value using a third loss function;
determining, based on the second training input image data and using a fourth portion of the first network, third representation data;
determining, based on the third representation data and the associated label data, a fifth loss value using a fourth loss function;
determining a third weight associated with the fourth loss value;
determining a fourth weight associated with the fifth loss value; and
determining the third loss value further based on the fourth loss value, the third weight, the fifth loss value, and the fourth weight.