Is there any method for threshold and face recognition to overcome with

Question

I am facing bug in flutter face dectection and recognition app where the it provide registerd face for new face which have no existing image but it match with other images.... while i am picking image from gallery for recognizing how to resolve these issue while i am get embedding okay but may something wrong with threshold setting or embedding normalization.

  List<int> _getEmbedding(Uint8List faceImageBytes) {
    final faceImage = img.decodeImage(faceImageBytes);
    if (faceImage == null) {
      log("Error decoding face image");
      return [];
    }
    // Preprocess the image to create a 4D input tensor.
    List<List<List<List<int>>>> input = _preprocessImage(faceImage)
        .map((e) => e
            .map((e) => e.map((e) => e.map((e) => e.toInt()).toList()).toList())
            .toList())
        .toList();

    // Get the output tensor shape from the interpreter.
    var outputShape = _interpreter.getOutputTensor(0).shape; // e.g. [1, 512]
    int outputLength = outputShape.reduce((a, b) => a * b);

    // Create the output container with the correct shape.
    var output = List.filled(outputLength, 0.0).reshape(outputShape);

    // Run inference.
    _interpreter.run(input, output);

    // Return the first (and only) row of the output.
    return List<int>.from(output[0]);
  }

  /// Recognize the provided face by comparing its embedding to registered faces.
  void recognizeFace() {
    if (embedding.isEmpty) {
      Get.snackbar("Error", "No face embedding computed.");
      return;
    }
    if (registeredFacesData.isEmpty) {
      Get.snackbar("Error", "No registered face data available.");
      return;
    }
    // double minDistance = 1.0;
    double minDistance = double.infinity;
    RegisteredFace? bestMatch;
    for (var face in registeredFacesData) {
      double distance = euclideanDistance(face.embedding, embedding);

      if (distance < minDistance) {
        minDistance = distance;
        bestMatch = face;
      }
    }
    // double threshold = 0.3; // Stricter threshold for accurate matching
    double threshold = 1.1;
    if (minDistance < threshold && bestMatch != null) {
      matchedRegisteredImageUrl.value = bestMatch.imageUrl;
      matchDistance.value = minDistance;
      Get.to(() => FaceRecognitionView());
    } else {
      matchedRegisteredImageUrl.value = "assets/no_match.png";
      matchDistance.value = minDistance;
      Get.snackbar(
        "No Match",
        "No matching face found (min distance: ${minDistance.toStringAsFixed(2)}).",
      );
    }
  }

 double euclideanDistance(List<double> emb1, List<double> emb2) {
    if (emb1.length != emb2.length) {
      throw ArgumentError('Embedding vectors must be of the same length');
    }
    double sum = 0.0;
    for (int i = 0; i < emb1.length; i++) {
      sum += math.pow(emb1[i] - emb2[i], 2);
    }
    return math.sqrt(sum);
  }

  List<double> _normalize(List<double> embedding) {
    double magnitude = math.sqrt(embedding.fold(0, (sum, e) => sum + e * e));
    return magnitude == 0
        ? embedding
        : embedding.map((e) => e / magnitude).toList();
  }

I am fixing the face recognition error it gives well result for faces which have existing face in backend while it also give a face image for unregistered face when I recognize it. but with some pics it also give sometime wrong face for as its existing face is available.