Guidelines for Liveness Detection

Image Type and Resolution

The size, type, and resolution of the image captured for liveness detection are important factors that affect its ability to be processed and analyzed accurately by the system.

Ensuring that the image is in the correct format and has sufficient resolution is necessary to obtain an accurate liveness detection result.

Note: Passive liveness supports two capture modes: single image and video input (Video Liveness). The specifications below differ by capture mode, but the face appearance and image quality guidance applies to both. See Passive Liveness for how the two modes differ.

Specifications - Single Image

Characteristic	Specification
Camera	Front camera of the device
Image Resolution	Minimum: 480x640 pixels Maximum: 3000x4000 pixels
Image Size	Minimum : 50KB Maximum : 1MB
Image Type	JPEG or JPG
Compression recommendation	Bicubic compression or resize the image resolution without affecting the image quality.
JPEG quality	No less than 70%
Image aspect ratio	Use the similarity of the initial ratio to reduce the size. Ratio 3:4
Selfie Orientation	Frontal face, portrait, straight to the camera and not rotated
Face area size	Minimum face area 200 x 200 pixels Maximum face area respective to the maximum allowed image file size Minimum Inter Eye Distance 160 pixels

Specifications - Video Liveness

The SDK produces a video that meets these limits automatically. They apply to any video submitted to the endpoint.

Characterisctic	Specification
Camera	Front camera of the device
Duration	5 seconds or less
File size	Under 50 MB
Resolution	Longest side under 4096 pixels
Orientation	Frontal face, portrait straight to the camera and not rotated

What Changes with Best Frame Capture

Video input does not remove the need for a good capture. It gives the system more than one frame to choose from, so a brief blur or a blink during the recording is less likely to cause a rejection.

The face appearance requirements are the same as for single image. The selected frame must contain one clearly visible, unobstructed, front-facing face. A video in which the face is never fully visible is still rejected.

Guideline	Single image	Best Frame Capture
Blur and motion	Rejected if the image is blurred	More tolerant. A sharper frame elsewhere in the video can be selected instead
Blinking and momentary expression	Rejected if the eyes are closed in that image	More tolerant. A frame with open eyes can be selected instead
Lighting and exposure	Rejected if the image is too dark or too bright	More tolerant, but only if lighting varies during the recording. Consistently poor lighting still causes rejection
Face occluded (mask, hat, glasses, hand, hair)	Rejected	No change
Multiple faces in frame	Rejected	No change. See Multiple faces in the frame
Face orientation and distance	Rejected if not front-facing or too small	No change if the condition persists for the whole recording
Enhanced or filtered capture	Rejected	No change

Important: All guidance on this page still applies to video input. Video input reduces rejections caused by momentary problems. It does not compensate for a poor environment, an obstructed face, or an incorrect capture position.

Capture Checklist

Image quality

• Make sure the capture is well-lit, and not too dark or too bright

• Avoid capturing in front of a bright light source

• Use a plain, neutral background with no distracting elements

• Avoid glare and reflections

• Avoid a blurred capture

• Use the original capture from the camera. Do not apply re-touch, beautification, or colour filters

• Do not add padding (black bars or any colored borders) to any side of the image. Padding is classified as image manipulation and will cause the image to be rejected.

Face appearance

• The entire face must be visible and not partially obscured by hair, glasses, a mask, a hat, a hand, or any other object

• Keep a neutral expression. Avoid smiling, frowning, or exaggerated expressions

• Keep the face straight-on and centred in the frame

• Keep only one face in the frame

• Do not hold a KTP or ID card in the selfie. This may be detected as spoof activity

Guidelines for Optimal Images

Image Condition	Acceptance/Rejection
Ideal	Highly recommended for optimal liveness detection accuracy
Relatively Safe	Acceptable, but may impact liveness detection accuracy
Avoid	Likely to be rejected. Significant impact on liveness detection accuracy

Guidelines for Image Quality

Image Condition	Acceptance/Rejection	Defective Image	Recommendation
Low Light	Warn be avoided as it can significantly impact liveness detection accuracy		The image should be well-lit and properly exposed
Blur	Warn be avoided as it can significantly impact liveness detection accuracy		The image should be sharp and in focus
Defect / Corrupt	Warn be avoided as it can significantly impact liveness detection accuracy		The image should be clean and undamaged
Over-exposure	Warn be avoided as it can significantly impact liveness detection accuracy		The image should be properly exposed with no blown-out highlights
Low Contrast	Warn be avoided as it can significantly impact liveness detection accuracy		The image should have good contrast with clearly defined edges
Glare	Warn be avoided as it can significantly impact liveness detection accuracy		The image should be free of glare and reflections
Enhanced Selfie	Warn be avoided as it can significantly impact liveness detection accuracy		The image should be original image taken from direct camera. Avoid applying any re-touch, beautification, color filter etc
Padding / Added Borders	Warn to be avoided as it can significantly impact liveness detection accuracy. Padding on any side of the image is classified as image manipulation.		The image should be the original capture with no padding, borders, or added margins on any side. Do not resize or reformat the image in a way that adds padding.

Guidelines for Face Appearance

Condition	Reason	Selfie should be avoid	Recommendation
Ideal Condition	Taking a selfie in proper lighting conditions to ensure the face is clear		NA
Ideal Condition	Using a hijab that is not covering the facial area		NA
Warn to be Avoided	Rotated Selfie in 90, 180, 270 degree		Suggest to take selfie in well tilt and in normal orientation
Relatively Safe	Head covered by hair bangs		Suggest to take selfie with no object covering the head/face
Relatively Safe	Wearing clear eyeglasses		Suggest to avoid the reflection in eyeglasses
Warn to be Avoided	Reflection in eyeglasses		Suggest to avoid the reflection in eyeglasses
Warn to be Avoided	Hair obstructing part of the face and eyes		Suggest retaking photo with hair pulled back
Warn to be Avoided	Face partially cut off in the image		Suggest retaking photo with entire face visible
Warn to be Avoided	Wearing colored cooling glasses that obscure the eyes		Suggest retaking photo without glasses or clear glasses
Warn to be Avoided	Wearing a mask that covers a significant portion of the face		Suggest retaking photo without a mask
Warn to be Avoided	Wearing a hat that obstructs the view of the eyes		Suggest retaking photo without a hat
Warn to be Avoided	Hand covering part of the face		Suggest retaking photo with hands away from face
Warn to be Avoided	Face not oriented properly or at an angle		Suggest retaking photo with face oriented straight ahead
Warn to be avoided	User holding KTP in the selfie		Suggest retaking photo without holding KTP
Warn to be Avoided	Multiple faces in the selfie		Suggest retaking photo only one face in the frame

Image Padding

Padding refers to any added border or margin — typically black, white, or colored bars — on one or more sides of the image. This commonly occurs when the image is resized, reformatted to a different aspect ratio, or processed through a third-party application before submission.

The liveness system treats any form of padding as image manipulation. Images with padding on any side — left, right, top, or bottom — and in any color will be rejected regardless of the face quality within the image.

This applies to both single image and video input capture modes.

Recommendations

• Submit the original image directly from the device camera without post-processing.

• Do not resize or reformat the image to a different aspect ratio, as this may introduce padding.

• If integrating via API, ensure that the image pipeline does not add borders, letterboxing, or canvas padding before submission.

• Confirm that any SDK or wrapper used in the capture flow preserves the original image dimensions without modification.

Multiple Faces in the Frame

The liveness check accepts one face per capture. If a second face is detected, the capture is rejected with an image quality error and the user is asked to keep only one face in the frame.

Only a clearly visible face counts as a second face. A person standing far from the camera, or a face only partly inside the frame, is normally not counted. This reduces failed captures in busy places such as a branch office, a kiosk, or a public area, where other people may pass behind the user.

Recommendations

• Capture in a position where no other person is inside the camera frame

• Move away from crowded areas, or turn so that other people are behind the camera instead of behind you

• If the error still appears, ask the user to check the camera preview before capturing

Note: Available from Android SDK 1.9.6+. This behaviour applies to both capture modes. With video input, the whole video is still rejected if a second face is present — video input does not make this check more tolerant.

If your use case requires strict one-person enforcement (for example, supervised onboarding where no other person may appear at all), contact the VIDA team.

Managing Liveness Errors & Recommendations

A person far behind the user is normally not counted as a second face from Android SDK 1.9.6.