This repo has been archived

Matching libraries are now distributed at Facial-Attribute-Detection-Android.

---

Face classifiers for Android

The project contains 3 classifiers:

1. Glasses/sunglasses – detect glasses and sunglasses on a face
2. Face covering – detect face being covered, e.g., by a mask
3. ID card face – detect whether a face image on an ID card is fake or genuine

Installation

Repository setup

The libraries are hosted in Applied Recognition's GitHub Packages repository.

Store your GitHub user name and token in your local.properties file as gpr.user and gpr.token respectively. Then in your build file add the following repository in your repositories block:

maven {
    url = uri("https://maven.pkg.github.com/AppliedRecognition/Ver-ID-Releases-Android")
    credentials {
        username = settings.extra["gpr.user"] as String?
        password = settings.extra["gpr.token"] as String?
    }
}

Dependencies to import

Glasses/sunglasses

com.appliedrec.verid3:face-classification-glasses:1.0.0

Face covering

com.appliedrec.verid3:face-classification-face-covering:1.0.0

ID card face

com.appliedrec.verid3:face-classification-id-card-face:1.0.0

Usage

First, you will need to detect a face in an image. For best results import the Ver-ID face detection library from com.appliedrec.verid3:face-landmark-detection-mp:1.0.0. You will also need to create an image type that can be used by Ver-ID. If your source image is an Android Bitmap you will want to import com.appliedrec.verid3:common-serialization:1.0.0.

Detect a face in image

Let's assume that you have an Android Bitmap in which you want to detect a face and run face classification.

import com.appliedrec.verid3.common.Face
import com.appliedrec.verid3.common.Image
import com.appliedrec.verid3.common.serialization.fromBitmap
import com.appliedrec.verid3.facedetection.mplandmarks.FaceDetection

suspend fun detectFaceInBitmap(context: Context, bitmap: Bitmap): Pair<Image,Face> {
    val faceDetection = FaceDetection(context)
    val image = Image.fromBitmap(bitmap)
    val face = faceDetection.detectFacesInImage(image, 1).first()
    return image to face
}

Run face classification

// Glasses detection
// The GlassesClassificationResult class has 2 properties:
// glasses: Float and sunglasses: Float. Each property can have a value between 0 and 1. If clear glasses 
// are detected the glasses property will be > 0.5. If sunglasses are detected both the glasses and sunglasses
// properties will have a value > 0.5.
suspend fun detectGlassesInFace(context: Context, face: Face, image: Image): GlassesClassificationResult {
    // Create an instance of the GlassesClassifier
    // The create factory method is suspending, you need to call it in a coroutine context
    val classifier = GlassesClassifier.create(context)
    // The classifier classes implement the AutoCloseable interface, either call the extractAttribute
    // method in a use block or call close() on the classifier once you're done with it. This ensures 
    // the classifier resources are cleaned up after use.
    return classifier.use {
        it.extractAttribute(face, image)
    }
}

// Face covering detection
// The face covering classifier returns a value between 0 and 1. 1 means highest confidence that
// the face is covered.
suspend fun detectFaceCoveringInFace(context: Context, face: Face, image: Image): Float {
    // Create an instance of the FaceCoveringClassifier
    // The create factory method is suspending, you need to call it in a coroutine context
    val classifier = FaceCoveringClassifier.create(context)
    // The classifier classes implement the AutoCloseable interface, either call the extractAttribute
    // method in a use block or call close() on the classifier once you're done with it. This ensures 
    // the classifier resources are cleaned up after use.
    return classifier.use {
        it.extractAttribute(face, image)
    }
}

// Fake ID image detection
// The classifier returns a value between 0 and 1. 1 means highest confidence that the face is 
// genuine part of the ID card. 0 means it's most likely tampered.
suspend fun detectGenuineIdCardFace(context: Context, face: Face, image: Image): Float {
    // Create an instance of the IdCardFaceClassifier
    // The create factory method is suspending, you need to call it in a coroutine context
    val classifier = IdCardFaceClassifier.create(context)
    // The classifier classes implement the AutoCloseable interface, either call the extractAttribute
    // method in a use block or call close() on the classifier once you're done with it. This ensures 
    // the classifier resources are cleaned up after use.
    return classifier.use {
        it.extractAttribute(face, image)
    }
}