Abstract: A head-mounted wearable device includes a frame mountable on a head of a user; an infrared imaging device arranged to image a face of the user when the frame is mounted on the head of the user; and a computing system configured to perform operations including causing the infrared imaging device to capture an image of the face of the user using infrared light received at the infrared camera and initiating a biometric authentication process based on the image. The head-mounted wearable device may include a visible-light imaging device to image the face of the user with the computing system configured to perform operations including causing the visible-light imaging device to capture a second image of the face of the user using visible light received at the visible-light imaging device, with the biometric authentication process being based in part on the second image.

Abstract: A method for detecting adverse conditions associated with a device includes receiving, at one or more processing devices at one or more locations, one or more image frames; receiving a set of signals representing outputs of one or more sensors of a device; estimating, based on the one or more image frames, a first set of one or more motion values; estimating, based on the set of signals, a second set of one or more motion values; determining that a degree of correlation between (i) a first motion represented by the first set of one or more motion values and (ii) a second motion represented by the second set of one or more motion values fails to satisfy a threshold condition; and in response to determining that the degree of correlation fails to satisfy the threshold condition, determining presence of an adverse condition associated with the device.

Abstract: This specification includes a method that includes receiving, at one or more processing devices at one or more locations, one or more image frames; receiving a set of signals representing outputs of one or more sensors of a device; estimating, based on the one or more image frames, a first set of one or more motion values; estimating, based on the set of signals, a second set of one or more motion values; determining that a degree of correlation between (i) a first motion represented by the first set of one or more motion values and (ii) a second motion represented by the second set of one or more motion values fails to satisfy a threshold condition; and in response to determining that the degree of correlation fails to satisfy the threshold condition, determining presence of an adverse condition associated with the device.

Abstract: Technology described herein can be embodied in a method for preventing access to a secure system based on determining a captured image to be of an alternative representation of a live person. The method includes capturing a first image and a second image of a subject illuminated by electromagnetic radiation in a first and a second wavelength ranges, respectively. The method also includes extracting, from the first image, a first portion representative of a sclera region of the subject, and from the second image, a second portion representative of the same region. It is determined that each of the first portion and the second portion includes features representative of vasculature in the sclera region, and in response, the subject in the image is identified to be an alternative representation of a live person. Upon search identification, the method includes preventing access to the secure system.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for biometric identity authentication. In one aspect, a method comprises: determining: (i) a plurality of first interest points in a first image of an eye of a first user, and (ii) a plurality of second interest points in a second image of an eye of a second user; determining a plurality of interest point pairs; processing a network input comprising the interest point pairs using an inlier neural network to generate a network output comprising a plurality of inlier scores; and determining whether an identity of the first user matches an identity of the second user based on the inlier scores.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training an encoder neural network having multiple encoder neural network parameters. The encoder neural network is configured to process a biometric data sample in accordance with current values of encoder neural network parameters to generate as output an embedded representation of the biometric data sample. The embedded representation includes: (i) an inter-class embedded representation, and (ii) an intra-class embedded representation that is different than the inter-class embedded representation.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining the identity of a user. In one aspect, a method comprises: obtaining a multi-spectral image that depicts an eye of a user, wherein the multi-spectral image comprises a plurality of registered two-dimensional channels, and each two-dimensional channel corresponds to a different spectrum of the multi-spectral image; processing the multi-spectral image using an encoder neural network to generate a fused image, wherein the fused image has a single two-dimensional channel; determining a set of features characterizing the eye of the user from the fused image; and determining an identity of the user based at least in part on the set of features characterizing the eye of the user.

Abstract: In a feature extraction and pattern matching system, image sharpening can enable vascular point detection (VPD) for detecting points of interest from visible vasculature of the eye. Pattern Histograms of Extended Multi-Radii Local Binary Patterns and/or Pattern Histograms of Extended Multi-Radii Center Symmetric Local Binary Patterns can provide description of portions of images surrounding a point of interest, and enrollment and verification templates can be generated using points detected via VPD and the corresponding descriptors. Inlier point pairs can be selected from the enrollment and verification templates, and a first match score indicating similarity of the two templates can be computed based on the number of inlier point pairs and one or more parameters of a transform selected by the inlier detection. A second match score can be computed by applying the selected transform, and either or both scores can be used to authenticate the user.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining the identity of a user. In one aspect, a method comprises: obtaining a multi-spectral image that depicts an eye of a user, wherein the multi-spectral image comprises a plurality of registered two-dimensional channels, and each two-dimensional channel corresponds to a different spectrum of the multi-spectral image; processing the multi-spectral image using an encoder neural network to generate a fused image, wherein the fused image has a single two-dimensional channel; determining a set of features characterizing the eye of the user from the fused image; and determining an identity of the user based at least in part on the set of features characterizing the eye of the user.

Abstract: Technology described herein can be embodied in a method for preventing access to a secure system based on determining a captured image to be of an alternative representation of a live person. The method includes capturing a first image and a second image of a subject illuminated by electromagnetic radiation in a first and a second wavelength ranges, respectively. The method also includes extracting, from the first image, a first portion representative of a sclera region of the subject, and from the second image, a second portion representative of the same region. It is determined that each of the first portion and the second portion includes features representative of vasculature in the sclera region, and in response, the subject in the image is identified to be an alternative representation of a live person. Upon search identification, the method includes preventing access to the secure system.

Abstract: In a feature extraction and pattern matching system, image sharpening can enable vascular point detection (VPD) for detecting points of interest from visible vasculature of the eye. Pattern Histograms of Extended Multi-Radii Local Binary Patterns and/or Pattern Histograms of Extended Multi-Radii Center Symmetric Local Binary Patterns can provide description of portions of images surrounding a point of interest, and enrollment and verification templates can be generated using points detected via VPD and the corresponding descriptors. Inlier point pairs can be selected from the enrollment and verification templates, and a first match score indicating similarity of the two templates can be computed based on the number of inlier point pairs and one or more parameters of a transform selected by the inlier detection. A second match score can be computed by applying the selected transform, and either or both scores can be used to authenticate the user.

Abstract: In a feature extraction and pattern matching system, image sharpening can enable vascular point detection (VPD) for detecting points of interest from visible vasculature of the eye. Pattern Histograms of Extended Multi-Radii Local Binary Patterns and/or Pattern Histograms of Extended Multi-Radii Center Symmetric Local Binary Patterns can provide description of portions of images surrounding a point of interest, and enrollment and verification templates can be generated using points detected via VPD and the corresponding descriptors. Inlier point pairs can be selected from the enrollment and verification templates, and a first match score indicating similarity of the two templates can be computed based on the number of inlier point pairs and one or more parameters of a transform selected by the inlier detection. A second match score can be computed by applying the selected transform, and either or both scores can be used to authenticate the user.

Abstract: In a feature extraction and pattern matching system, image sharpening can enable vascular point detection (VPD) for detecting points of interest from visible vasculature of the eye. Pattern Histograms of Extended Multi-Radii Local Binary Patterns and/or Pattern Histograms of Extended Multi-Radii Center Symmetric Local Binary Patterns can provide description of portions of images surrounding a point of interest, and enrollment and verification templates can be generated using points detected via VPD and the corresponding descriptors. Inlier point pairs can be selected from the enrollment and verification templates, and a first match score indicating similarity of the two templates can be computed based on the number of inlier point pairs and one or more parameters of a transform selected by the inlier detection. A second match score can be computed by applying the selected transform, and either or both scores can be used to authenticate the user.

Abstract: In a feature extraction and pattern matching system, image sharpening can enable vascular point detection (VPD) for detecting points of interest from visible vasculature of the eye. Pattern Histograms of Extended Multi-Radii Local Binary Patterns and/or Pattern Histograms of Extended Multi-Radii Center Symmetric Local Binary Patterns can provide description of portions of images surrounding a point of interest, and enrollment and verification templates can be generated using points detected via VPD and the corresponding descriptors. Inlier point pairs can be selected from the enrollment and verification templates, and a first match score indicating similarity of the two templates can be computed based on the number of inlier point pairs and one or more parameters of a transform selected by the Inlier detection. A second match score can be computed by applying the selected transform, and either or both scores can be used to authenticate the user.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training an encoder neural network having multiple encoder neural network parameters. The encoder neural network is configured to process a biometric data sample in accordance with current values of encoder neural network parameters to generate as output an embedded representation of the biometric data sample. The embedded representation includes: (i) an inter-class embedded representation, and (ii) an intra-class embedded representation that is different than the inter-class embedded representation.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for biometric identity authentication. In one aspect, a method comprises: determining: (i) a plurality of first interest points in a first image of an eye of a first user, and (ii) a plurality of second interest points in a second image of an eye of a second user; determining a plurality of interest point pairs; processing a network input comprising the interest point pairs using an inlier neural network to generate a network output comprising a plurality of inlier scores; and determining whether an identity of the first user matches an identity of the second user based on the inlier scores.

Abstract: Biometric enrollment and verification techniques for ocular-vascular, periocular, and facial regions are described. Periocular image regions can be defined based on the dimensions of an ocular region identified in an image of a facial region. Feature descriptors can be generated for interest points in the ocular and periocular regions using a combination of patterned histogram feature descriptors. Quality metrics for the regions can be determined based on region value scores calculated based on texture surrounding the interest points. A biometric matching process for calculating a match score based on the ocular and periocular regions can progressively include additional periocular regions to obtain a greater match confidence.

Abstract: Methods and systems for securing biometric templates and generating secret keys are provided. One or more images are received. Interest points are identified based on the received images, and a plurality of obfuscating data points are generated based on the interest points. An obfuscated template based on the interest points and the obfuscating data points is created and stored. A secret key can be encoded using a subset of at least one of the obfuscating data points and the interest points in the template.

Abstract: This specification describes technologies relating to biometric authentication based on images of the eye. In general, one aspect can be embodied in methods that include obtaining one or more image regions from a first image of an eye. Each of the image regions may include a view of a respective portion of the white of the eye. The method may further include applying several distinct filters to each of the image regions to generate a plurality of respective descriptors for the region. The several distinct filters may include convolutional filters that are each configured to describe one or more aspects of an eye vasculature and in combination describe a visible eye vasculature in a feature space. A match score may be determined based on the generated descriptors and based on one or more descriptors associated with a second image of eye vasculature.

Abstract: This specification describes technologies relating to biometric authentication based on images of the eye. In general, one aspect of the subject matter described in this specification can be embodied in methods that include obtaining a first image of an eye including a view of the white of the eye. The method may further include determining metrics for the first image, including a first metric for reflecting an extent of one or more connected structures in the first image that represents a morphology of eye vasculature and a second metric for comparing the extent of eye vasculature detected across different color components in the first image. A quality score may be determined based on the metrics for the first image. The first image may be rejected or accepted based on the quality score.