Changelog¶
das-Face v3.24¶
- This new release updates the antispoofing model used for passive and active Veridas Liveness Solutions, maintaining the same conversion rate, but improving the detection of presentation attacks.
- We improved the conversion rate of
verification/video
endpoint.
das-Face v3.23¶
Changed¶
- This update replaces all liveness detection methods with a new model that successfully passed iBeta's evaluations for both level 1 and level 2 security. This enhances our anti-spoofing measures for a more secure experience.
das-Face v3.22¶
Changed¶
- A new face detector was added to the endpoint
authenticity/photo
.
das-Face v3.21¶
- This release includes an improved face detector in
/v2/verification/photo
endpoint that increases the conversion rate:- In
selfie-mode
operations by 0.3 pp. - In
document-mode
from 0.5 pp (native
) up to 3.0 pp (web-desktop
).
- In
das-Face v3.20¶
- This release allows the generation of biometric credentials using a new lightweight biometric model whose target is deploying facial authentication when there are low computational resources.
- The new model, which is only available for endpoints
/v2/models/{hash}/...
and/v2/verification/credential
, is identified by:- Model hash:
7c8e92b813baf87f8ec56dd840c32b4294ba14ddda436065f26ec7a0
- Model tag:
20230403
- Model hash:
- The new model, which is only available for endpoints
- Deprecation announcement: The
rotatePhotos
property ofverification/photo
endpoint is marked as deprecated, enforcing deprecation for the 2024.5 release.
das-Face v3.19¶
- This release multiplies by 4 the overall antispoofing accuracy of our passive liveness and Selfie Alive Pro (SAP) solutions by increasing security for the same conversion rates.
- This release includes an improved face detector that increases the
conversion rate for selfie-vs-document use case when generating
biometric credentials in
/v2/models/{hash}/{mode}/credential/photo
.
das-Face v3.18¶
- Added a refined image validation feature that specifically targets the quality and framing of submitted selfies.
- Only On-Prem: the RAM requirement can be reduced drastically by customizing the models and/or endpoints to be used at runtime.
das-Face v3.14¶
- Improved the user experience by introducing a new calibration for passive-selfie in both the
web-desktop
andweb-mobile
environments.- These improvements are expected to significantly increase the conversion rate to ~94% while maintaining a high level of security, ensuring that users can authenticate themselves more easily and efficiently.
- Added new endpoint to the API
/v2/authenticity/video/photo
, which computes the authenticity of a video record and compares a given face photo to the face of the person recorded in the video. The conversion rate of this endpoint is ~96%. More details in the product documentation.
das-Face v3.13¶
- The measured conversion of Selfie Alive Pro, the liveness technology that is designed to detect and prevent spoofing attacks, has increased by 12 points on the desktop environment.
das-Face v3.12¶
- The new biometric model has been designed to increase the overall accuracy of the system by reducing false positive and false negative rates. The FPR of selfie comparison (known as
selfie-mode
) has been reduced by 27%, from 0.0060% @FNR=0.171 to 0.0044% @FNR=0.165 at 0.8 threshold. Similarly, the FPR of the comparison of a selfie photo and an identity document photograph (also known asdocument-mode
) has improved by 77%, from 0.27% @FNR=1.57% at 0.7 threshold to 0.061% @FNR=1.204%.- The new model is identified by:
- Model hash:
2e0cb1f8dd755e2c9caf18c48c2b550dc7e64e57205b82c3fc5c62d6
- Model tag:
20221017
- Model hash:
- The new model is identified by:
- Additionally, the processing time for photo verification has improved by 25%, decreasing from 872ms to 652ms (endpoint:
/v2/verification/photo
).
- Deprecation enforcement. The endpoint
credential/photo
has been deprecated. From now on, it will be necessary to use the endpointmodels/{hash}/{mode}/credential/photo.
Enforcement of deprecation for the model identified with:
- Tag:
20190813
- Hash:
3a9e9d5ffd5de4c212c2aff26eeca523fb69754e604894520b32e4ed
Existing credentials generated with such model won’t be compatible anymore with das-Face.
das-Face v3.10¶
- This release improves the active liveness antispoofing performance, in Selfie Alive Pro (SAP). In this case, the BPCER (%) is 30.64% and ACPER 2.02%. In other words, assuming the worst-case scenario of Level 2 attacks, 30.64% of authentic cases are rejected and 2.02% of spoofing attempts will be misclassified as genuine. This supposes a reduction in the APCER of almost 60% comparing it to the previous version.
- Deprecation reminder: As announced in version v3.6, the
credential/photo
endpoint is marked as deprecated, enforcing deprecation for the 2022 Q4 release. Keep in mind that after the 2022 Q4 release, it will be necessary to use the endpointmodels/{hash}/{mode}/credential/photo.
- Deprecation reminder: As announced in version v3.6, the biometric model with model hash
904fa9ef6e71ef541f20a95d3dc97821b7af43b8cd2c1bb3eb09df15
and model tag20200514
is marked as deprecated, enforcing deprecation for the 2022 Q4 release. Existing credentials generated with that model have to be re-generated before the 2022 Q4 release with one of the supported models to continue being functional. Keep in mind that after the 2022 Q4 release, existing credentials generated with that model won’t work anymore.
das-Face v3.9¶
- This release improves the passive liveness antispoofing performance by increasing the system security. This means an improvement in the ACPER, which
is considered a measure about system security.
- Regarding passive liveness detection, using an operation point at 0.70, the 3.56% of authentic cases will be rejected and the 0.6% of spoofing attempts will be misclassified as authentic.
- Deprecation reminder: As announced in version v3.6, the
credential/photo
endpoint is marked as deprecated, enforcing deprecation for the 2022 Q4 release. Keep in mind that after the 2022 Q4 release, it will be necessary to use the endpointmodels/{hash}/{mode}/credential/photo.
- Deprecation reminder: As announced in version v3.6, the biometric model with model hash
904fa9ef6e71ef541f20a95d3dc97821b7af43b8cd2c1bb3eb09df15
and model tag20200514
is marked as deprecated, enforcing deprecation for the 2022 Q4 release. Existing credentials generated with that model have to be re-generated before the 2022 Q4 release with one of the supported models to continue being functional. Keep in mind that after the 2022 Q4 release, existing credentials generated with that model won’t work anymore.
das-Face v3.8¶
- This release updates the usage tracking library.
- Deprecation reminder: As announced in version v3.6, the
credential/photo
endpoint is marked as deprecated, enforcing deprecation for the 2022 Q4 release. Keep in mind that after the 2022 Q4 release, it will be necessary to use the endpointmodels/{hash}/{mode}/credential/photo.
- Deprecation reminder: As announced in version v3.6, the biometric model with model hash
904fa9ef6e71ef541f20a95d3dc97821b7af43b8cd2c1bb3eb09df15
and model tag20200514
is marked as deprecated, enforcing deprecation for the 2022 Q4 release. Existing credentials generated with that model have to be re-generated before the 2022 Q4 release with one of the supported models to continue being functional. Keep in mind that after the 2022 Q4 release, existing credentials generated with that model won’t work anymore.
das-Face v3.7¶
- das-Face integrates a new Deepfake Detection Algorithm into Veridas liveness detection systems. A Deepfake attack, also known as Face or Identity Swap, is a type of impersonation attempt through an injection attack, where the attacker injects a pre-recorded or live video of the attacker where the victim's face replaces the attacker's face. The Deepfake Detection Algorithm works with stills and videos and is combined with the rest of Veridas' anti-spoofing technologies, giving an overall confidence score for liveness. The current DeepFake detection technology has a maximum APCER of 15.31% for a BPCER of 0.31%.
- The involved endpoints are:
/v2/authenticity/photo
/v2/challenges/analysis/video-photo
- The overall processing time has been increased on
/v2/authenticity/photo
by less than a 25% (from 595 ms to 743 ms) and on/v2/challenges/analysis/video-photo
by less than a 12% (from 3458 ms to 3859 ms).
- The involved endpoints are:
- Deprecation reminder: As announced in version v3.6, the
credential/photo
endpoint is marked as deprecated, enforcing deprecation for the 2022 Q4 release. Keep in mind that after the 2022 Q4 release, it will be necessary to use the endpointmodels/{hash}/{mode}/credential/photo.
- Deprecation reminder: As announced in version v3.6, the biometric model with model hash
904fa9ef6e71ef541f20a95d3dc97821b7af43b8cd2c1bb3eb09df15
and model tag20200514
is marked as deprecated, enforcing deprecation for the 2022 Q4 release. Existing credentials generated with that model have to be re-generated before the 2022 Q4 release with one of the supported models to continue being functional. Keep in mind that after the 2022 Q4 release, existing credentials generated with that model won’t work anymore.
das-Face v3.6¶
- This version adds improvements in anti-spoofing technology both for passive and active technology. The new algorithm improves detection accuracy and has passed the evaluation tested by iBeta to the ISO 30107-3 Biometric Presentation Attack Detection Standard, and the PAD Level 2 Confirmation Letter is here.
Passive Liveness detection¶
For passive liveness detection, the improvement is shown in the next table (threshold of 0.70):
Max APCER | dasFace 3.5.2 BPCER = 3% | dasFace 3.6.0 BPCER ≈ 1.5% |
---|---|---|
ibeta level 1 | 8.82 % | 3.67 % |
If the main goal is protecting from Level 2 attacks, it is encouraged to increase the threshold to 0.95.
Max APCER | dasFace 3.5.2 BPCER ≈ 25.93% | dasFace 3.6.0 BPCER ≈ 28.60% |
---|---|---|
ibeta level 1 | 0.4202 % | < 0.63 % |
ibeta level 2 | 23.32 % | 7.861 % |
Selfie Alive Pro - Active Liveness detection¶
For activeliveness detection, the improvement is shown in the next table (threshold of 0.70):
Max APCER | dasFace 3.5.2 BPCER = 5% | dasFace 3.6.0 BPCER = 5% |
---|---|---|
ibeta level 1 | 25.00 % | 1.176 % |
If the main goal is protecting from Level 2 attacks, it is encouraged to increase the threshold to 0.90.
Max APCER | dasFace 3.5.2 BPCER = 15% | dasFace 3.6.0 BPCER = 31% |
---|---|---|
ibeta level 1 | 1.052 % | < 1.18 % |
ibeta level 2 | 48.69 % | 5.05 % |
- Deprecation announcement: The
credential/photo
endpoint is marked as deprecated, enforcing deprecation for the 2022 Q4 release. Keep in mind that after the 2022 Q4 release, it will be necessary to use the endpointmodels/{hash}/{mode}/credential/photo.
- Deprecation announcement: The biometric model with model hash
904fa9ef6e71ef541f20a95d3dc97821b7af43b8cd2c1bb3eb09df15
and model tag20200514
is marked as deprecated, enforcing deprecation for the 2022 Q4 release. Existing credentials generated with that model have to be re-generated before the 2022 Q4 release with one of the supported models to continue being functional. Keep in mind that after the 2022 Q4 release, existing credentials generated with that model won’t work anymore.
das-Face v3.5¶
- Improved passive
antispoofing
performance.- On printed mask attacks, for a threshold of 0.7, the ACPER goes from 0.0126 to 0.0042, improving a 66.66% with respect to the previous release. BPCER has no measurable differences.
- Reduced the number of
0.0
cases on Active antispoofing (SelfieAlivePro), analyzing previously discarded cases and providing insight to the user. - Added procedure for image turn (0, 90,180, 270 degrees) on endpoint
verification/photo
. With the propertyrotatePhotos
, the endpoint performs a rotate operation until a face is detected successfully. This can be applied foranchor
,target
orboth
images. - Improved error handling on
dasface
. - Processing time has been reduced a 9% on average.
das-Face v3.4¶
- New biometric model which reduces false positive and false negative rates, and improves the accuracy on different demographic cases (Caucasian, African, Indian, Asian).
- The FNR of selfie-mode has been reduced a 15%, from 0.20% to 0.17% at 0.8 threshold, and the funnel of document-mode is improved by almost one point maintaining the same security level, obtaining FNR=1.57% @ FPR=0.27% at 0.7 threshold.
- This model has obtained a reduction of the error up to 55% with respect to the previous model in NIST evaluations (
veridas-007
). - The new model is identified by:
- Model hash:
1f29f3d61aecf88b961d33bad141a98d9da514c8bac7c2b81eb94235
- Model tag:
20210913
- Model hash:
- Processing time has been reduced a 15% on average.
- Active antispoofing (SelfieAlivePro) funnel has been increased a 10%.
- Endpoint
challenges/analysis/video-photo
now also supports frame arrays plus audio file. This implies a simpler integration with iOS HTML SDK on versions prior to 15.0.
das-Face v3.3 (2021-Q3)¶
- Improved active
antispoofing
performance on HTML.- For a
threshold
of 0.7 BPCER is 82% lower with respect to the previous release. APCER has no measurable differences - For a
threshold
of 0.9 BPCER is 80% lower with respect to the previous release. APCER has no measurable differences.
- For a
Deprecation enforcement
Enforcement of deprecation for the model identified with:
- Tag:
20190813
- Hash:
3a9e9d5ffd5de4c212c2aff26eeca523fb69754e604894520b32e4ed
Existing credentials generated with such model won’t be compatible anymore with das-Face.
das-Face v3.2 (2021-Q2)¶
- Passive
antispoofing
performance has been improved.- For a
threshold
of 0.7 APCER is 46% lower with respect to the previous release. BPCER has no measurable differences - For a
threshold
of 0.9 APCER is 57% lower with respect to the previous release. BPCER has no measurable differences.
- For a
verification/photo
average processing time improved 22,48% from the previous release.challenges/analysis/video-photo
average processing time improved 8% from the previous re- lease.- Unique ID will be returned to the VeriSaaS API users using the
X-Request-Id
header.
das-Face v3.0 (2021-Q1)¶
This version comes with two major changes related to biometric models and antispoofing techniques.
- New biometric model which reduces false positive and false negative rates, and improves the accuracy on different demographic cases (Caucasian, African, Indian, Asian). The model improves the accuracy of selfie-mode from 99.8% to 99.9% and the funnel of document-mode is improved by two points maintaining the same security level, obtaining TPR=2.43% @ FPR=0.62% at 0.7 threshold. The new model is identified by:
- Model hash:
904fa9ef6e71ef541f20a95d3dc97821b7af43b8cd2c1bb3eb09df15
- Model tag:20210203
- Because model 20210203 generates larger biometric credentials than earlier models, the default endpoint /credential/photo will generate them using model 20200514, so by default credentials have the 2896 bits length as usual.
-
When security is a must, and having a use case that could afford using larger biometric credentials, the model 20210203 is available at the endpoint:
/models/<hash>/<mode>/credential/photo
The biometric credentials for model 20210203 are of 9040 bits length.
- An update of the SelfieAlive Pro (SAP) use case (/challenges/analysis/video-photo) decreasing the time of the operation in a 26%. The system is calibrated at threshold 0.7 for BPCER=2.7% and APCER=0.9%.
- An update of the passive AS solution (/authenticity/photo) calibrated at threshold 0.7 for BPCER=3.2% and APCER=7.8%, reducing a 50% the APCER of the previous das-Face ver- sion. The system requires additional 300ms per query (a total of 1185ms per query).
- Deprecation announcement: The biometric model with model hash
3a9e9d5ffd5de4c212c2aff26eeca523fb69754e604894520b32e4ed
and model tag20190813
is marked as deprecated, enforcing deprecation for the 2021 Q3 release. Existing credentials generated with that model have to be re-generated before the 2021 Q3 release with one of the supported models to continue being functional. Keep in mind that after the 2021 Q3 release, existing credentials generated with that model won’t work anymore. Remember that there is available a new endpoint to check the model used for creating the credentials/models/metadata/from-credential
.
das-Face v2.7.2 (2021-Q1)¶
- Fixes low scores at
/verification/video
when using mode=document-mode.
das-Face v2.9 (2021-Q1)¶
- Add support for BBVA SEMaaS traceId field (using the x-rho-traceid header).
das-Face v2.8 (2021-Q1)¶
- Added support for JPEG blobs.
das-Face v2.7 (2020-Q4)¶
This new version of das-Face comes with new liveness detection capabilities.
- Improved passive liveness detection with the following figures:
- Improves replay-attacks APCER from 27% in das-Face v2.5.2 to 5.3% in das-Fave v2.6.7.
- Achieves BPCER=11.7%
- These operational figures are valid for selfie-alive and vali-Das on-boarding products.
- Adds Selfie-Alive Pro use case, VERIDAS liveness detection using an active procedure.
- The system performs with APCER=0.1% and BPCER=12.9%.
- Supports videos recording a sequence of random head movements.
- This solution is in compliance to ISO 30107-3 PAD level 1, as tested by iBeta.
- Support for a contextual data header:
X-Veridas-RTag
header is logged in every query allowing to build statistics and billing segregated by customer, use cases, sub-clients and so on.
Deprecation enforcement
Enforcement of deprecation for the model identified with:
- Tag:
20180827
- Hash:
b0bf475e5344e816f12b83c13c075a3256ef95e60ac1cdc273aef59f
Existing credentials generated with such model won’t be compatible any more with das-Face.
das-Face v2.6 (never released)¶
This version was used internally to calibrate the liveness detection released in das-Face v2.7.
das-Face v2.5 (2020-mid-Q3)¶
- Fix bug in
v2/verification/video
when more than one face was detected in the video.
das-Face v2.4 (2020-Q2)¶
- New biometric model which reduces false positive and false negative rates, and improves the accuracy on different demographic cases (Caucasian, African, Indian, Asian). The model improves accuracy of selfie-mode form 99.6% to 99.8% and document-mode from 97.87% to 98.34%. The new model is identified by:
- Model hash:
904fa9ef6e71ef541f20a95d3dc97821b7af43b8cd2c1bb3eb09df15
- Model tag:
20200514
- Added a new endpoint to retrieve the operational model from the biometric credential meta- data, i.e., model hash and operation mode chosen to generate the given credential.
- Added a new endpoint to get the operational models active in das-Face.
- Added a description of the orchestration required for selfie-alive.
- Deprecation announcement The biometric model with:
- model hash:
b0bf475e5344e816f12b83c13c075a3256ef95e60ac1cdc273aef59f
- model tag:
20180827
is marked as deprecated, enforcing deprecation for 2020 Q4 release. Existing credentials generated with that model have to be re-generated before 2020 Q4 release with one of the supported models to continue being functional. Keep in mind that after 2020 Q4 release, existing credentials generated with that model won’t work anymore. Remember that from this release there is available a new endpoint to check the model used for creating the credentials.
das-Face v2.3 (2020-Q1)¶
- The endpoint POST
/v2/authenticity/photo
(antispoofing) accepts images images taken with Veridas HTML SDKs. - Deprecation reminder: API v1 endpoints are not supported any more.
das-Face v2.2 (2019-Q4)¶
- Improved log messages, increasing their exploitability.
- Recall deprecation of API v1 endpoints, it will be enforced at 2020 Q1.
das-Face v2.1 (2019-Q3)¶
- Improved document-mode biometric model for selfie-vs-document functionality.
- Improved selfie-mode biometric model for selfie-vs-selfie functionality.
- Improved anti-spoofing detection for photo authenticity endpoint, increasing accuracy from 90.3% to 92.2% for replay-attacks.
- Replacement of dataset used to compute photo authenticity performance curve (anti-spoofing). The new replay-attack dataset is larger, uses more devices and has been carried out under more complex conditions.
- Notice that due to the new biometric model for document-mode and selfie-mode, all endpoints are expected to use the new model.
- Recall the deprecation of API v1 endpoints. We encourage our customers to start using the new API v2 because API v1 will be removed at 2020 Q1.
das-Face v2.0 (2019-Q2)¶
- New API v2 with better semantics, syntax and improved error handling.
- New endpoint
/v2/credential/photo
to generate a biometric credential from an image. The returned credential contains the biometric data of the face found in the image. - New endpoint
/v2/verification/credential
to compare the image of a face with a previously generated biometric credential. - Deprecation of API v1 endpoints. We encourage migration to API v2 as soon as possible.