Create demo.java

models/face_recognition_sface/demo.java

@@ -0,0 +1,121 @@
+import nu.pattern.OpenCV;
+import org.opencv.core.Mat;
+import org.opencv.core.Size;
+import org.opencv.imgcodecs.Imgcodecs;
+import org.opencv.objdetect.FaceDetectorYN;
+import org.opencv.objdetect.FaceRecognizerSF;
+
+/**
+ * the java demo of FaceRecognizerSF
+ *
+ * you need the dependencies of maven：
+ *
+ *   *OpenCV Java bindings packaged with native libraries,
+ *   seamlessly delivered as a turn-key Maven dependency.
+ *   You don't need to download OpenCV to support the relevant features.
+ *   doc:https://github.com/openpnp/opencv?tab=readme-ov-file
+ *
+ *   https://mvnrepository.com/artifact/org.openpnp/opencv
+ *   <dependency>
+ *       <groupId>org.openpnp</groupId>
+ *       <artifactId>opencv</artifactId>
+ *       <version>4.9.0-0</version>
+ *   </dependency>
+ */
+public class FaceRecognizer {
+
+    private static double cosine_similar_threshold = 0.363;
+
+    private static double l2norm_similar_threshold = 1.128;
+
+    // Your full path of yunet model
+    // <a href="https://docs.opencv.org/4.8.0/df/d20/classcv_1_1FaceDetectorYN.html">FaceDetectorYN</a>
+    private static String faceDetectModelPath = "/face_detection_yunet_2023mar.onnx";
+    // state faceDetector
+    private static FaceDetectorYN faceDetector = null;
+
+    // Your full path of sface model
+    private static String faceRecognizModelPath = "/face_recognition_sface_2021dec.onnx";
+    // state faceRecognizer
+    private static FaceRecognizerSF faceRecognizer = null;
+
+    public static void main(String[] args) {
+        // You need to use the full path of img please
+        boolean b = faceRecognizer("imgPathA", "imgPathB");
+        System.out.println(b);
+    }
+
+    public static boolean faceRecognizer(String imgPathA, String imgPathB) {
+        // Load for opencv
+        OpenCV.loadLocally();
+        // Load for faceDetector
+        loadFaceDetector();
+        // Load for faceRecognizer
+        loadFaceRecognizer();
+
+        return faceRecognizerUtil(imgPathA, imgPathB);
+    }
+
+    // Load of faceDetector
+    private static void loadFaceDetector() {
+        if (faceDetector != null) {
+            return;
+        }
+        // You could use the full path for faceDetect model instead to get the resource
+        faceDetector = FaceDetectorYN.create(faceDetectModelPath, "", new Size());
+    }
+
+    // Load for faceRecognizer
+    private static void loadFaceRecognizer() {
+        if (faceRecognizer != null) {
+            return;
+        }
+        // You could use the full path for faceRecogniz model instead to get the resource
+        faceRecognizer = FaceRecognizerSF.create(faceRecognizModelPath, "");
+    }
+
+    /**
+     * FaceRecogniz. Calculating the distance between two face features
+     *
+     * @param imgPathA the path of imgA
+     * @param imgPathB the path of imgB
+     */
+    private static boolean faceRecognizerUtil(String imgPathA, String imgPathB) {
+        // 1.Read img convert to a mat
+        Mat imgA = Imgcodecs.imread(imgPathA);
+        Mat imgB = Imgcodecs.imread(imgPathB);
+
+        // 2.Detect face from given image
+        Mat faceA = new Mat();
+        faceDetector.setInputSize(imgA.size());
+        faceDetector.detect(imgA, faceA);
+        Mat faceB = new Mat();
+        faceDetector.setInputSize(imgB.size());
+        faceDetector.detect(imgB, faceB);
+
+        // 3.Aligning image to put face on the standard position
+        Mat alignFaceA = new Mat();
+        faceRecognizer.alignCrop(imgA, faceA.row(0), alignFaceA);
+        Mat alignFaceB = new Mat();
+        faceRecognizer.alignCrop(imgB, faceB.row(0), alignFaceB);
+
+        // 4.Extracting face feature from aligned image
+        Mat featureA = new Mat();
+        faceRecognizer.feature(alignFaceA, featureA);
+        featureA = featureA.clone();
+        Mat featureB = new Mat();
+        faceRecognizer.feature(alignFaceB, featureB);
+        featureB = featureB.clone();
+
+        // 5.FaceRecogniz. Calculating the distance between two face features. If the condition is met, it returns true
+        // Get cosine similar
+        double match1 = faceRecognizer.match(featureA, featureB, FaceRecognizerSF.FR_COSINE);
+        // Get l2norm similar
+        double match2 = faceRecognizer.match(featureA, featureB, FaceRecognizerSF.FR_NORM_L2);
+        if (match1 >= cosine_similar_threshold && match2 <= l2norm_similar_threshold) {
+            return true;
+        } else {
+            return false;
+        }
+    }
+}