Biometric Anonymizer Systems and Methods

By performing multidimensional vector encoding, normalization, and binning permutation key processing on biometric data, anonymized records are generated, solving legal and privacy issues when sharing biometric data and achieving accurate matching and computational efficiency of anonymized data.

CN123076867APending Publication Date: 2026-11-10ARTYN INCYTE SOLUTIONS
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202580027223.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2024-03-22
Filing Date
2025-03-20
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing technologies make it difficult to share biometric data without violating laws and privacy regulations, while maintaining data anonymization and matching accuracy, and existing anonymization methods may lead to the risk of recovering the original PII.

Method used

By performing multidimensional vector encoding on biometric data, and applying normal distribution normalization, binning, and permutation key techniques, anonymized records are generated, while maintaining the homomorphic relationship of the data during the anonymization process to ensure matching accuracy.

Benefits of technology

It enables the sharing of anonymized data among different entities without leaking PII, maintaining matching accuracy, reducing the possibility of recovering the original PII, and reducing computational complexity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123076867A_ABST
    Figure CN123076867A_ABST
Patent Text Reader

Abstract

In one aspect, this disclosure provides a system and method for anonymizing data, which may include: receiving biometric data having personally identifiable information associated with a person; encoding a record with a plurality of biometric identifiers of the biometric data, each of the plurality of biometric identifiers including a numerical value; and generating an anonymized record based on transforming the numerical value of each of the plurality of biometric identifiers, the anonymized record being deassociated from the personally identifiable information. Various aspects of this disclosure further provide methods for comparing and matching anonymized data.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] Cross-references to related applications

[0002] This application claims priority to U.S. Provisional Application No. 63 / 568,690, filed March 22, 2024, the entire contents of which are incorporated herein by reference.

[0003] field

[0004] This disclosure generally relates to anonymizing data, and more specifically to anonymizing biometric data to remove personally identifiable information, and even more specifically to anonymizing biometric data to remove personally identifiable information while maintaining homomorphism with the original data.

[0005] background

[0006] Personal biometric information may be subject to strict terms of use. Privacy regulations define such use as for authorized purposes. Furthermore, biometric information is often retained without the direct consent of the individual concerned. And, regardless of whether consent is given, obligations related to the retention and use of biometric information are increasing, and depending on the jurisdiction involved, breaches of these obligations often result in significant, and usually punitive, consequences.

[0007] In many cases, balancing the obligation to manage custody with respecting the boundaries of authorized purposes is problematic, especially given the evolving nature of privacy regulations and case history. Organizations processing personally identifiable information (PII) face the ongoing challenge of correctly asserting the following: when does their right to retain PII data end, and which data pursuits fall within the scope of the authorized purposes? Alternatively, organizations may take an overly conservative stance, i.e., not using available data to support their efforts, particularly when there are doubts about their consistency with the approved purposes or the right to retain data.

[0008] Therefore, it remains desirable to develop further improvements and advancements in the collection, maintenance, sharing, and / or anonymization of biometric information and / or PII to overcome the shortcomings of known technologies and provide additional advantages.

[0009] This section aims to introduce various aspects of the art that may be associated with this disclosure. Such discussion is intended to help provide a benchmark for reference to facilitate a better understanding of specific aspects of this disclosure. Accordingly, it should be understood that this section should be interpreted in this light, and not necessarily as an admission that it is prior art. Attached Figure Description

[0010] Embodiments of this disclosure will now be described by way of example only with reference to the accompanying drawings.

[0011] Figure 1Systems and methods for encoding and matching biometric information according to embodiments of this disclosure are shown.

[0012] Figure 2 Systems and methods for encoding, anonymizing, and matching biometric information are illustrated according to embodiments of this disclosure.

[0013] Figure 3 An example of applying binning to a normal distribution according to an embodiment of this disclosure is shown.

[0014] Figure 4 Systems and methods for applying new permutation keys, according to embodiments of this disclosure, are illustrated.

[0015] Figure 5 It is a block diagram of an example computing device or system for implementing one or more systems, aspects, embodiments, methods or operations of this disclosure.

[0016] In all accompanying figures, sometimes only one instance or fewer of the elements visible in the view are designated by guide lines and figure labels, simply for simplicity and to avoid confusion. However, it should be understood that in such cases, all other instances are similarly designated and covered by the corresponding description.

[0017] Detailed description

[0018] In one aspect, this disclosure provides a computer-implemented method for anonymizing data, the method comprising: receiving biometric data having personally identifiable information associated with a person; encoding a record with a plurality of biometric identifiers of the biometric data, each of the plurality of biometric identifiers including a numerical value; and generating an anonymized record based on a transformation of the numerical value of each of the plurality of biometric identifiers, the anonymized record being deassociated from the personally identifiable information.

[0019] As a further example of this disclosure, transforming the value includes normalizing the value for each of the plurality of biometric identifiers.

[0020] As a further example of what is being disclosed, normalization includes scaling the value to a normal distribution curve of a plurality of representative samples, each of which has a plurality of corresponding biometric identifiers.

[0021] As a further example of what is being disclosed, normalization involves scaling and shifting the value to a normal distribution curve of a plurality of representative samples, each of which has a plurality of corresponding biometric identifiers.

[0022] As a further example of this disclosure, the normal distribution curve is based on the average mean of the means of the plurality of representative samples and the average mean of the standard deviations of the plurality of representative samples.

[0023] As a further example of what is being disclosed, transforming the value involves applying a many-to-one mapping to that value.

[0024] As a further example of this disclosure, transforming the value includes assigning the value of each of the plurality of biometric identifiers to a representative value of a bin in the plurality of bins.

[0025] As a further example in accordance with this disclosure, the multiple boxes comprise at least 80 boxes.

[0026] As a further example of this disclosure, transforming the value includes rearranging the order of the plurality of biometric identifiers in the encoded record to a first order different from the original order.

[0027] As a further example in accordance with this disclosure, rearranging the order includes applying a first permutation key.

[0028] Further examples of this disclosure include: restoring the original order of the plurality of biometric identifiers in the encoded record based on the inverse of the first permutation key, and rearranging the order of the plurality of biometric identifiers in the encoded record into a second order different from the first order based on the application of a second permutation key.

[0029] Further examples of this disclosure include: maintaining the records of the plurality of biometric identifiers of the coded record in the first order until the plurality of biometric identifiers of the coded record are rearranged into the second order.

[0030] As a further example of the anonymized record according to this disclosure, the anonymized record includes a first anonymized record having a first plurality of biometric identifiers, each of the first plurality of biometric identifiers including a first value, the method further including: receiving a second anonymized record including a second plurality of biometric identifiers, each of the second plurality of biometric identifiers including a second value; and determining a match between the first anonymized record and the second anonymized record based on comparing the first value and the second value of each of the first plurality of biometric identifiers and the second plurality of biometric identifiers, wherein the match includes a confidence interval and / or a confidence level.

[0031] As a further example of this disclosure, determining the match includes assessing the proximity between the first value and the second value.

[0032] In one aspect, this disclosure provides a system for anonymizing data, the system comprising: a biometric device for obtaining biometric information associated with a person; and a processor configured to perform a method according to this disclosure.

[0033] In one aspect, this disclosure provides a computer-readable medium having instructions stored thereon for anonymizing data, which, when executed by a processor, cause the processor to perform a method according to this disclosure.

[0034] To facilitate understanding of the principles of this disclosure, reference may be made to the features illustrated in the accompanying drawings, and these features will be described using specific language. However, it should be understood that this is not intended to limit the scope of this disclosure. Any changes and further modifications, as well as any further applications of the principles of this disclosure as described herein, are contemplated as would normally be expected by one of ordinary skill in the art to which this disclosure pertains. It will be apparent to one of ordinary skill in the art that some features not related to this disclosure may not be shown in the accompanying drawings for clarity.

[0035] First, for ease of reference, certain terms used in this application and their meanings in the context are explained below. Where a term used herein is not defined below, the broadest definition given by one of skill in the art, such as that reflected in at least one printed publication or published patent, should be given. Furthermore, this process is not limited to the use of the terms shown below, as all equivalents, synonyms, new developments, and terms or processes serving the same or similar purposes are considered to be within the scope of this disclosure.

[0036] Personally identifiable information (PII) includes information representations that allow for a reasonable inference, directly or indirectly, of the identity of the individual to whom the information applies. Regulations may further define a company's responsibilities regarding the processing, protection, and sharing of PII.

[0037] Biometric identifiers include unique and / or measurable characteristics used to mark, describe, and identify an individual, and may include biometric information such as physical measurements and calculations that are quantitatively related to human characteristics. Biometric identifiers can be categorized into physiological and behavioral characteristics that produce identification or partial identification of an individual. Physiological characteristics, for example, relate to the geometry or measurement of shape, movement (e.g., gait), movement derivatives (e.g., signature), voice (e.g., speech), chemistry (e.g., biological sample testing), biometric signatures (e.g., iris scans, DNA, fingerprints, or palm prints), structure (e.g., medical imaging), body shape, or body parts (e.g., hands or ears). Biometric identifiers containing PIIs can be referred to as, for example, PII biometric data.

[0038] To address questions such as what percentage of a sample population exhibits a biometric trait, it's necessary to test members of the sample population against that biometric trait to separate matches from non-matches. When processing alphanumeric PII data (such as name, address, gender, age, or date of birth), matches and non-matches can be established using simple binary operators (such as equal to, less than, greater than, etc.). For example, age = X or year of birth between A and B. Even when applying advanced preprocessing techniques to such alphanumeric PII data, such as data correction, normalization, or binning (e.g., postal addresses often require data correction and / or normalization, or matching ages by comparing members of the sample population against age ranges), the end result remains the same: data matches and non-matches can be achieved using simple operators. However, biometric information may require matching using non-simple operators, such as measuring "proximity" or determining confidence levels. Consider the example of two facial images of an individual: a first image taken at a first time and a second image taken at a second time. The second image is unlikely to match the first image pixel-by-pixel, thus limiting the utility of simple binary operators in determining matches. The occurrence of mismatched pixels can be caused by, but is not limited to: changes in lighting conditions, changes in shooting angle, differences in facial expressions, or positional shifts in objects and faces due to changes in image size. Despite these differences, if the faces found in two images are "close enough" in terms of biometrics, the two images can be "matched".

[0039] An example of determining the closeness of biometric data for matching involves determining the cumulative variance across multiple measurements. After processing the measurements of two biometric samples, the biometric matching algorithm will not return a Boolean value of "yes, this matches" or "no, this doesn't match," but rather a confidence level (derived from the degree of the overall 'variance' between the two biometric samples). In the case of face recognition, the confidence level could correspond to the probability that two facial images belong to the same person.

[0040] In this embodiment, confidence levels can be mapped to confidence intervals. For example, multiple confidence intervals can be defined, such as "match," "partial match," and "no match," each associated with a confidence level range. For instance, a first confidence interval may correspond to a first confidence level range, a second confidence interval may correspond to a second confidence level range, a third confidence interval may correspond to a third confidence level range, and so on. In this embodiment, the first confidence interval is "match," and the first confidence level range is greater than or equal to 91%; the second confidence interval is "partial match," and the second confidence level range is less than 91% but greater than or equal to 85%; and the third confidence interval is "no match," and the third confidence level range is less than 85%. Confidence levels and confidence intervals, and their interpretation, typically reflect the underlying Biometric Sample Processing System (BSPS).

[0041] Figure 1 Examples of systems and procedures for matching biometric information, particularly for facial recognition, are shown.

[0042] Step 1 may include capturing a biometric sample 12 (e.g., a facial image, voice recording, fingerprint scan, etc.) using a biometric capture device 10, such as a camera for capturing images of a person's face. Step 1 may further include sending the captured biometric sample 12 (e.g., jpeg, wav, data file, etc.) to a recognition and selection engine 20 for integrity and plausibility checks, and separating relevant portions of the biometric sample 12. In embodiments, step 1 may further apply enhancements to the captured biometric sample 12 to, for example, improve the clarity of subsequent steps.

[0043] Step 2 may include sending the enhanced biometric sample 12' to the encoder 30.

[0044] Step 3 may include creating a first k-dimensional (kd) vector encoding (E1) representing a biometric sample, the first vector having k measurements corresponding to each feature of the biometric sample.

[0045] Step 4 may include retrieving or finding a previously stored vector code (E2) and further forwarding E1 and / or E2 to matcher 50. In an embodiment, matcher 50 may also find and / or retrieve one or more coded vectors. In an embodiment, step 4 may further include adding E1 to a database 40 of previously stored vector codes for known biometric samples.

[0046] Step 5 may include determining the proximity of the match between E1 and E2. In an embodiment, determining the proximity of the match includes measuring the Euclidean distance between E1 and E2. Step 5 may further include providing an output including a confidence level and / or a confidence interval.

[0047] according to Figure 1 Potential drawbacks of such systems include situations where custody rights to E1 and E2 belong to different entities. Legal and privacy regulations may restrict the sharing of PIIs by an entity having custody rights to E1 and vice versa, thus precluding the possibility of determining a match between E1 and E2 without violating legal and / or privacy regulations. Therefore, for this reason, and to relieve the entity with custody rights to E1 of obligations, it is necessary to remove or otherwise anonymize the PII to allow and permit the sharing of biometric information. This raises the technical challenge of balancing the anonymization of the PII with maintaining sufficient fidelity to perform further functions (such as matching the proximity of anonymized PII data).

[0048] Anonymizing biometric information, including PIIs, may require processing one or more dimensions of the data. For example, anonymizing data can benefit from normalizing each feature of an encoded k-dimensional vector to a normal distribution curve to marginalize or eliminate discriminative differences between the distributions of each encoded feature. As another example, applying a many-to-one mapping to each feature of an encoded k-dimensional vector can marginalize or eliminate the ability to inversely map a feature back to its original value. As yet another example, applying permutations to the encoded vector to rearrange the order of features in the encoded k-dimensional vector can provide a higher degree of anonymity to the data. Typically, each of the aforementioned examples removes the ability to recover the PII by increasing the fuzziness or anonymity of the underlying data, while maintaining homomorphism with the original data and allowing matching with other similarly anonymized data at a level of accuracy comparable to matching using the original data.

[0049] The advantages achieved by this disclosure include, but are not limited to: anonymizing the PII to the extent that the anonymized data does not allow for the recovery, discovery, or reproduction of the original PII; providing real-time and / or near-real-time processing equivalent to the closeness of matching the original PII; achieving measurable matching accuracy comparable to the accuracy and matching performance when operating on the original PII, with low false positives and low false negatives; and enabling data sharing among different entities without disclosing the PII.

[0050] The systems and methods described in this disclosure can derive anonymized forms of PIIs that can be matched with similarly anonymized PIIs for use in biometric applications, including but not limited to: facial recognition, fingerprint recognition, voice recognition, retinal recognition, DNA matching, and handwriting recognition. Various aspects of this document therefore provide anonymized data without PIIs (from which PII data cannot be recovered) for matching biometric identifiers without retaining the PIIs. In embodiments, PII anonymization occurs at or shortly after PII capture. Example embodiments of this disclosure may be applicable where: each biometric sample comprises a set of k values; the proximity of a matched anonymized sample is correlated with these k values; and the k-dimensional encoding has a number of dimensions that makes a brute-force attempt using k factorials computationally infeasible. In embodiments, when matching proximity of anonymized data, each of the k values ​​is applied with equal weighting.

[0051] This disclosure provides systems and methods for converting human-associated biometric information into a derived form without a Personal Indicator (PII). The PII-filtered or anonymized form of biometric information obscures the identifiable person associated with the original biometric information. However, anonymizing biometric information can allow for determining, for example, what percentage of a sample population exhibits biometric characteristics, or what percentage of a sample population has the same biometric identifier, such as the presence of a facial image in two different population samples.

[0052] This disclosure provides systems and methods that allow organizations with custody of PIIs to perform one or more of the following actions: use biometric information for additional purposes (globally, the PII is limited to the purpose for which it was collected); share PII anonymized biometric information with affiliates (once the biometric data no longer contains the PII, it is no longer subject to the same terms as the PII); retain PII anonymized biometric information after the point at which the original purpose has been fulfilled (and after the PII must be removed); remove the PII before the original purpose has been fulfilled, for example, as soon as the PII is captured and converted into PII anonymized biometric information; limit the retention of the PII to specific, justifiable circumstances (and then subject those circumstances to more stringent access control), and reduce the burden of complying with the responsible custody of the PII (which may also lead to higher expectations for more responsible action).

[0053] Anonymizing PIIs according to this disclosure may include: anonymizing PII data such that the PII cannot be recovered from the anonymized form; preserving the characteristics of the biometric information, thereby allowing for proximity matching in the anonymized form without a decrease in accuracy compared to matching using the original PII data; and matching the anonymized data without causing an excessive increase in computation time or resources. In embodiments, the systems and methods disclosed herein may include an expected complexity of O(N).

[0054] Figure 2 This demonstrates an example of anonymizing first data captured by a first party (A) to match it with second data obtained by a second party (B). For example... Figure 2 As shown, multiple steps allow data to be shared between two different parties without exposing any underlying PII to each other. For example, the first party (A) could be a first company, and the second party (B) could be a database 40' outside of the first party (A), such as a third-party cloud service.

[0055] Step 1 may include a first party (A) capturing a biometric sample 12 (e.g., a facial image, voice recording, fingerprint scan, etc.) using a biometric capture device 10, such as a camera for capturing images of a person's face. Step 1 may further include sending the captured biometric sample 12 (e.g., jpeg, wav, data file, etc.) to a recognition and selection engine 20 for integrity and plausibility checks, and separating relevant portions of the biometric sample 12. In embodiments, step 1 may further apply enhancements to the captured biometric sample to, for example, improve the clarity of subsequent steps.

[0056] Step 2 may include sending the enhanced biometric sample 12' to the encoder 30.

[0057] Step 3 may include creating a first k-dimensional (kd) vector encoding (E1) representing the biometric sample 12, which has k measurements for each feature of the biometric sample.

[0058] Step 3.5 may include anonymizing E1 using an anonymizer 35 according to embodiments of this disclosure.

[0059] Step 4 may include retrieving or finding a previously stored and anonymized vector code (E2), and further forwarding the anonymized E1 and / or E2 to the matcher 50'. In an embodiment, the matcher 50' may find and / or retrieve one or more anonymized coded vectors. In an embodiment, step 4 may further include adding the anonymized E1 to a database 40' of previously stored anonymized vector codes for known biometric samples.

[0060] Step 5 may include determining the proximity of the match between the anonymized E1 and E2. In an embodiment, determining the proximity of the match includes measuring the Euclidean distance between E1 and E2. Step 5 may further include providing an output that includes a confidence level and / or a confidence interval.

[0061] According to this disclosure, anonymizing data to prevent the recovery of PIIs involves applying a series of transformations to the biometric information containing the PII. Embodiments of this disclosure anonymize data in a manner that prevents PII recovery while maintaining homomorphism with the original data and allowing for real-time or near-real-time matching of proximity to other anonymized data. For example, using anonymized data does not result in a meaningful increase in the computational workload of proximity calculations compared to performing proximity calculations on the same original data before anonymization.

[0062] Consider an example of matching anonymized biometric information associated with facial recognition that has 128 features (i.e., k = 128).

[0063] As a preliminary step, embodiments of this disclosure may first collect a sample set of facial images of a crowd to determine the statistical properties of the sample set (e.g., the mean and standard deviation of various facial features used to perform facial recognition) for further matching with anonymized biometric information. For each sample image, a representation of 128 biometric identifiers or features f is generated. i The 128-d encoding. For 128 features f i Each of them determines the feature f on a sample set of facial images. i Calculated mean µ i and standard deviation σ i Approximate values; and calculate all 128 µ i average mean µ avr and all 128 σ i average mean σ avr .

[0064] In the first step, the method according to this disclosure can acquire, capture, obtain, or retrieve biometric information containing a person's PII, such as by capturing an image of a person's face using a camera or otherwise receiving and / or retrieving an image of a person's face. The first step may further include defining a bounding box around the captured face in the image and returning n box coordinates of the image file.

[0065] In the second step, the image can be encoded into a k-dimensional vector (E1), such as a 128-dimensional vector, according to the method disclosed herein, with each dimension corresponding to a biometric identifier for facial recognition.

[0066] In the third step, the method disclosed herein can anonymize the k-dimensional vector (E1) by applying a series of transformations to E1. Typically, each transformation increases the difficulty of recovering the PII while maintaining sufficient fidelity to compute the proximity of the match to other similarly anonymized data.

[0067] In an embodiment, the first transformation may include each feature f of E1. i Normalization can be performed. For example, since the characteristic distribution of each feature (e.g., the mean and standard deviation of a normal distribution) can help identify which feature each location represents, reducing the likelihood of recovering the PII can further include transforming and / or shifting the distribution of each feature so that the standard normal distribution has different values, such as a mean of 0 and a standard deviation of 1. For example, for each feature f i A transformation is applied to the location on the standard normal distribution. In an embodiment, the transformation includes:

[0068] Normalized f i Value = (original f) i Value - µ i ) / σ i

[0069] Normalization f of the standard normal distribution i The value can be further transformed or scaled to have a mean of µ by applying another transformation. avr and standard deviation σ avr The normal distribution, for example:

[0070] Scaling normalization f i Value = (normalized f) i Value - µ avr ) / σ avr

[0071] In the feature f i After normalization and transformation, even when examining a large number of samples of the representation, each feature will have the same distribution as all other features when considering the distribution, thus reducing and / or eliminating the possibility of distinguishing each feature based on its distribution, since each feature will have the same normalized normal distribution. Advantageously, determining the Euclidean distance between two such normalized vectors can maintain a lossless matching function relative to the original baseline.

[0072] In an embodiment, the second transformation may include applying a many-to-one mapping to the k-dimensional vector, such as through binning, which can act as a blurring of the PII, similar to blurring or pixelating an image to remove identifiable features. In the first step, binning may include selecting the number of bins, β. The lower the number of bins, the greater the anonymization or "blurring" effect. However, too few bins may negatively impact the accuracy of matching the proximity of anonymized data.

[0073] In the second step, binning may include dividing the scaled range of the normal distribution into β regions. In an embodiment, each of the β regions includes an equal probability, in other words, an equal area under the given normal distribution curve. Thus, regions closer to the curve's mean / middle will have a relatively smaller width than regions closer to the curve's tail. Figure 3 An example of a binning scheme based on this disclosure is shown.

[0074] In this embodiment, the boundary of each of the β regions can be defined using a Z-critical value equal to the following probabilities: 1 / β, 2 / β, ..., (β-1) / β. Given Z(pr) as a function of the Z-critical value used to calculate the probability pr, the β regions (each with an equal probability) will be:

[0075] {[ -∞ < x ≤ Z(1 / β)], [ Z(1 / β) < x ≤ Z(2 / β)], … [ Z((β -1) / β) < x ≤+∞]}.

[0076] In the third step, binning may include calculating the value of each bin [Val]. left < x ≤ Val right The representative value is Z((Prob). In an embodiment, the representative value is the midpoint of the bin range. In an embodiment, the representative value is set to Z((Prob). left +Prob right ) / 2), where Prob n Is the value less than Val? n The probability of.

[0077] In the fourth step, binning may include sorting each of the k-dimensional features into its corresponding bin and replacing feature f with the representative value of the corresponding bin. i The value of . In this respect, binning serves to provide an irreversible many-to-one mapping: each feature f has a value within a given bin range. i They will be mapped to the same representative value; therefore, the representative value can represent a range of values. Thus, the k-dimensional vector E1 at k = 128 can include a set of values ​​{b}. v1 , b v2 , … b v128 These values ​​reflect representative values ​​of their associated bins and can be used to determine the proximity of a match to similar binned data, where the accuracy of the match depends on the number of bins (as discussed). As the number of bins β increases, feature f... i The difference between the value of f and the representative value of the bin may decrease because each bin has a smaller width as the number of bins increases. Therefore, as the number of bins increases, the representative value of a given bin converges closer to the feature f.i The given value. In the embodiment, the number of boxes β is greater than or equal to 80.

[0078] In an embodiment, the third transformation may include anonymizing the representation of the values, for example, shuffling the original, normalized, and / or binned values ​​of E1. In an embodiment, anonymizing the representation of the values ​​includes applying permutations to the normalized and binned values, resulting in a reordering of the values. For example, a 128-d vector E1 includes 128! permutations (more than 3.8 × 10⁻⁶). 125 This presents several possibilities, making it computationally infeasible to unlock the correct permutation using brute force. Combined with normalization, even if the second party has access to a large number of samples of the permuted anonymized representations from the first party, the distribution values ​​of each of the 128 features will be identical, and the second party will be unable to discern which of the actual 128 features each anonymized feature represents. In an embodiment, the permutation is chosen and known by the first party and is not shared with the second party. In an embodiment, the permutation includes a permutation key.

[0079] Figure 4 An example of changing the permutation key is illustrated. A first party (A) captures a person's biometric information, including their PII (Personal Information Identity). According to this disclosure, the first party further anonymizes the biometric information before sending it to a second party (B). In this regard, the first party selects a permutation key to apply to all biometric samples during the anonymization step. However, at some point in the future, the first party may wish to apply a different permutation key.

[0080] In the first step, the first party may submit a request 60 to the key update engine 70 to change from the first replacement key to the second replacement key.

[0081] In the second step, the key update engine 70 can request all records anonymized with a given permutation key from a party outside the first party, such as requesting all records anonymized with the first permutation key stored in database 40' by the second party (B).

[0082] In the third step, anonymizer 35 receives records anonymized with the first permutation key for mapping the records to their representative order (e.g., the first value is f7, the second value is f). 89 Anonymizer 35 can further run the inverse of the permutation key on each record, undoing the reordering generated by the original application of the first permutation key, thereby retrieving the original feature order of each encoding ({f1, f2, …, f…). 128Anonymizer 35 can then further apply a second permutation key to replace the new feature order and further supply the newly reordered records to the key update engine. Anonymizer 35 can further store the second permutation key for use in anonymizing future biometric samples.

[0083] In the fourth step, the key update engine can supply the newly reordered records to the second party (B), which have been permuted based on the second permutation key. The two sets of records (Set1, the original record encoded with the original first permutation key, and Set2, the record encoded with the new second permutation key) can coexist for an indeterminate grace period. This allows for a zero-disruption transition period, during which new search requests (based on either key) will take effect. Once the transition period is complete (i.e., all users of the first key have transitioned to using the new permutation key), Set1 can be deleted.

[0084] As disclosed in this paper, the results of applying a series of transformations to the biometric PII produce anonymized data in which the PII cannot be recovered, and in which the anonymized data shares homomorphic relations with the original data and allows matching between other similar anonymized data without drastically increasing the computational load.

[0085] Figure 5 This is a block diagram of an example computerized device or system 500, which can be used to implement one or more aspects, components, sub-components, operations, etc. of an embodiment of a biometric anonymizer according to this disclosure, including embodiments of a method for anonymizing biometric data.

[0086] Computerized system 500 may include one or more of processor 502, memory 504, mass storage device 510, input / output (I / O) interface 506, and communication subsystem 508. Further, system 500 may include multiple, such as multiple processors 502 and / or multiple memories 504. Processor 502 may include one or more of digital processors, analog processors, digital circuitry designed for processing information, analog circuitry designed for processing information, state machines, and / or other mechanisms for electronically processing information. These processing units may be physically located within the same device, or processor 502 may represent the processing functions of multiple devices operating collaboratively. Processor 502 may be configured to execute modules, or otherwise perform functions attributed to modules, via software; hardware; firmware; a combination of software, hardware, and / or firmware; and / or other mechanisms for configuring processing capabilities on processor 502, and may include one or more physical processors (during execution of processor-readable instructions), circuitry, hardware, storage media, or any other components.

[0087] One or more of the components or subsystems of the computerized system 500 may be interconnected via one or more buses 512 or in any other suitable manner.

[0088] Bus 512 can be one or more of any type of bus architecture, including memory bus, storage device bus, memory controller bus, peripheral bus, etc. CPU 502 can include any type of electronic data processor. Memory 504 can include any type of system memory, such as dynamic random access memory (DRAM), static random access memory (SRAM), synchronous DRAM (SDRAM), read-only memory (ROM), combinations thereof, etc. In an embodiment, the memory can include ROM for use at startup and DRAM for data storage of programs and for use when executing programs.

[0089] Mass storage device 510 may include any type of storage device configured to store data, programs, and other information and make the data, programs, and other information accessible via bus 512. Mass storage device 510 may include one or more of solid-state drives, hard disk drives, disk drives, optical disk drives, etc. In some embodiments, data, programs, or other information may be stored remotely, such as in the cloud. Computerized system 500 may send information to or receive information from remote storage in any suitable manner, including via a network or other data communication medium through communication subsystem 508.

[0090] I / O interface 506 can provide an interface for enabling wired and / or wireless communication between computerized system 500 and one or more other devices or systems. For example, I / O interface 506 can be used for communicative coupling with sensors such as cameras or camcorders. Furthermore, additional or fewer interfaces can be utilized. For example, one or more serial interfaces, such as Universal Serial Bus (USB) (not shown), can be provided.

[0091] The computerized system 500 can be used to configure, operate, control, monitor, sense and / or adjust devices, systems and / or methods according to this disclosure.

[0092] A communication subsystem 508 may be provided for transmitting and receiving signals, or both, over digital data communications (including communication networks) of any form or medium. Examples of communication networks include local area networks (LANs), wide area networks (WANs), interconnected networks (e.g., the Internet), and peer-to-peer networks (e.g., self-organizing peer-to-peer networks). The communication subsystem 508 may include any component or set of components for enabling communication via one or more wired and wireless interfaces. These interfaces may include, but are not limited to, USB, Ethernet (e.g., IEEE 802.3), High Definition Multimedia Interface (HDMI), Firewire™ (e.g., IEEE 1394), Thunderbolt™, WiFi™ (e.g., IEEE 802.11), WiMAX (e.g., IEEE 802.16), Bluetooth™ or Near Field Communication (NFC), and GPRS, UMTS, LTE, LTE-A, and Dedicated Short Range Communication (DSRC). The communication subsystem 508 may include one or more ports or other components (not shown) for one or more wired connections. Alternatively or additionally, the communication subsystem 508 may include one or more transmitters, receivers, and / or antenna elements (all not shown).

[0093] Figure 5 The computerized system 500 is merely an example and is not intended to be limiting. Various embodiments may utilize some or all of the components shown or described. Some embodiments may use other components not shown or described but known to those skilled in the art.

[0094] In the foregoing description, numerous details have been set forth to provide a thorough understanding of the embodiments for the purpose of explanation. However, it will be apparent to those skilled in the art that these specific details are not essential. In other instances, well-known electrical structures and circuits have been illustrated in block diagram form to avoid obscuring the meaning. For example, no specific details are provided regarding whether the embodiments described herein are implemented as software routines, hardware circuits, firmware, or a combination thereof.

[0095] The embodiments of this disclosure can be represented as a computer program product stored in a machine-readable medium (also referred to as a computer-readable medium, processor-readable medium, or computer-usable medium having computer-readable program code embodied therein). A machine-readable medium can be any suitable tangible, non-transitory medium, including magnetic, optical, or electrical storage media (including magnetic disks, optical disc read-only memories (CD-ROMs), memory devices (volatile or non-volatile), or similar storage mechanisms)). The machine-readable medium can contain various sets of instructions, sequences of code, configuration information, or other data that, when executed, cause a processor to perform the steps in the methods according to the embodiments of this disclosure. Those skilled in the art will understand that other instructions and operations necessary to implement the described embodiments can also be stored on the machine-readable medium. Instructions stored on a machine-readable medium can be executed by a processor or other suitable processing device and can be interfaced with a circuit system to perform the described tasks.

[0096] The above embodiments are intended to be illustrative only. Changes, modifications, and alterations can be made to specific embodiments by those skilled in the art. The scope of the claims should not be limited to the specific embodiments set forth herein, but should be interpreted in a manner consistent with the overall specification.

Claims

1. A computer-implemented method for anonymizing data, the method comprising: Receive biometric data containing personally identifiable information associated with a person; The record is encoded using multiple biometric identifiers of the biometric data, each of which includes a numerical value and... Anonymous records are generated by transforming the values ​​of each of the plurality of biometric identifiers, and the anonymized records are decoupled from the personally identifiable information.

2. The method as described in claim 1, wherein, Transforming the numerical values ​​includes normalizing the numerical values ​​of each of the plurality of biometric identifiers.

3. The method as described in claim 2, wherein, Normalization involves scaling the values ​​to a normal distribution curve of multiple representative samples, each of which has multiple corresponding biometric identifiers.

4. The method of claim 2, wherein, Normalization involves scaling and shifting the numerical values ​​to a normal distribution curve of multiple representative samples, each of which has multiple corresponding biometric identifiers.

5. The method as described in claim 3 or claim 4, characterized in that, The normal distribution curve is based on the average mean of the means of the multiple representative samples and the average mean of the standard deviations of the multiple representative samples.

6. The method of claim 1, wherein, Transforming the value involves applying a many-to-one mapping to the value.

7. The method of claim 1, wherein, Transforming the values ​​includes assigning the values ​​of each of the plurality of biometric identifiers to representative values ​​of bins in the plurality of bins.

8. The method of claim 7, wherein, The plurality of boxes includes at least 80 boxes.

9. The method of claim 1, wherein, Transforming the numerical value includes rearranging the order of the plurality of biometric identifiers in the encoded record to a first order that is different from the original order.

10. The method of claim 9, wherein, Rearranging the order includes applying a first permutation key.

11. The method of claim 10, further comprising: Based on the inverse of the first permutation key, the original order of the plurality of biometric identifiers in the encoded record is restored, and Based on the application of the second permutation key, the order of the plurality of biometric identifiers in the encoded record is rearranged into a second order that is different from the first order.

12. The method of claim 11, further comprising: The records of the plurality of biometric identifiers in the encoded record are maintained in the first order until the plurality of biometric identifiers in the encoded record are rearranged into the second order.

13. The method of claim 1, wherein, The anonymized record includes a first anonymized record having a first plurality of biometric identifiers, each of the first plurality of biometric identifiers including a first numerical value, and the method further includes: Receive a second anonymized record comprising a second plurality of biometric identifiers, each of the second plurality of biometric identifiers including a second numerical value, and Based on comparing the first value and the second value of each of the first plurality of biometric identifiers and the second plurality of biometric identifiers, a match is determined between the first anonymized record and the second anonymized record; The matching includes confidence intervals and / or confidence levels.

14. The method of claim 13, wherein, Determining the match involves assessing the proximity between the first value and the second value.

15. A system comprising: A biometric device, said biometric device being used to obtain biometric information associated with a person, and A processor configured to perform the method as described in any one of claims 1 to 14.

16. A computer-readable medium having instructions stored thereon, which, when executed by a processor, cause the processor to perform the method as described in any one of claims 1 to 14.