faceVerificationUtilization/utils.py

import cv2 as cv
import numpy as np
from PIL import Image
import torch
import pandas as pd
from .setConfig import efficientnet_model, face_detector, transform, pca_xgb, faiss, load_db
import os
import requests

def ImgPreprocessing(img):
    if len(img.shape) == 2 or img.shape[2] == 1:
        img = cv.cvtColor(img, cv.COLOR_GRAY2BGR)

    img_yuv = cv.cvtColor(img, cv.COLOR_BGR2YUV)
    clahe = cv.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
    img_yuv[:, :, 0] = clahe.apply(img_yuv[:, :, 0])
    img = cv.cvtColor(img_yuv, cv.COLOR_YUV2BGR)
    return img

def YoloFaceDetection(img):
    results = face_detector.predict(img, conf=0.7)

    keyReturn = {
        'message': "",
        'status': False,
        'coordinate': []
    }

    if len(results) == 0 or results[0].boxes is None or len(results[0].boxes) == 0:
        keyReturn['message'] = "Tidak ada wajah terdeteksi"
        return keyReturn

    boxes = results[0].boxes.xyxy.cpu().numpy()
    confs = results[0].boxes.conf.cpu().numpy()

    max_idx = confs.argmax()
    x1, y1, x2, y2 = map(int, boxes[max_idx])

    keyReturn['message'] = "Face detected"
    keyReturn['status'] = True
    keyReturn['coordinate'] = [x1, y1, x2, y2]

    return keyReturn

def FaceValidationPredict(**face_crop):
    x1, y1, x2, y2 = face_crop['coordinate']
    face_crop = face_crop['img'][y1:y2, x1:x2]

    face_crop = cv.cvtColor(face_crop, cv.COLOR_BGR2GRAY)
    face_crop = cv.cvtColor(face_crop, cv.COLOR_GRAY2RGB)  
    face_pil = Image.fromarray(face_crop)

    face_tensor = transform(face_pil).unsqueeze(0)

    with torch.no_grad():
        features = efficientnet_model(face_tensor).cpu().numpy()

    pred = pca_xgb.predict(features)[0]

    return pred, features, face_crop


# feature
def FaceValidation(frame: np.ndarray):
    if frame is None:
        return "No frame captured from webcam"
    
    if isinstance(frame, dict):
        frame = frame['image']

    pred = ''
    img = ImgPreprocessing(frame)
    keyReturn = YoloFaceDetection(img)

    if keyReturn['status'] is False:
        return keyReturn['message']

    results = keyReturn['coordinate']
    x1, y1, x2, y2 = results

    validation, features, face_crop = FaceValidationPredict(coordinate=[x1, y1, x2, y2], img=img)
    pred = 'Wajah Valid' if validation == 0 else 'Wajah Tidak Valid'

    return f"Predicted class: {pred}"

def FaceRecord(img, name, photo_idx):
    if img is None or name is None:
        return f"Foto {photo_idx}] Gagal: Tidak ada gambar atau nama"

    os.makedirs('users', exist_ok=True)
    user_dir = os.path.join("users", name)
    os.makedirs(user_dir, exist_ok=True)

    img = np.array(img)
    img = ImgPreprocessing(img)
    keyReturn = YoloFaceDetection(img)
    if keyReturn['status'] is False:
        return keyReturn['message']

    x1, y1, x2, y2 = keyReturn['coordinate']
    pred, features, face_crop = FaceValidationPredict(coordinate=[x1, y1, x2, y2], img=img)

    if pred == 1:  
        return f"[Foto {photo_idx}] Gagal: wajah tidak valid"

    save_path = os.path.join(user_dir, f"photo_{photo_idx}.jpg")
    cv.imwrite(save_path, face_crop)

    csv_path = "users/face_features.csv"
    row = pd.DataFrame({
        "label": [name],
        "features": [features.flatten().tolist()]  
    })

    if not os.path.exists(csv_path):
        row.to_csv(csv_path, index=False)
    else:
        row.to_csv(csv_path, mode="a", index=False, header=False)

    total_fotos = len([f for f in os.listdir(user_dir) if f.endswith(".jpg")])
    if total_fotos < 4:
        return f"[Foto {photo_idx}] Berhasil disimpan ke {save_path}. ({total_fotos}/4)"
    else:
        return f"[Foto {photo_idx}] Berhasil disimpan ke {save_path}. ✅ Semua 4 foto sudah lengkap!"

def Recognize(frame: np.ndarray):
    if frame is None:
        return "No frame captured from webcam"
    
    if isinstance(frame, dict):
        frame = frame['image']

    img = ImgPreprocessing(frame)
    keyReturn = YoloFaceDetection(img)

    if not keyReturn['status']:
        return keyReturn['message']

    results = keyReturn['coordinate']
    x1, y1, x2, y2 = results
    pred, features, face_crop = FaceValidationPredict(coordinate=[x1, y1, x2, y2], img=img)
    faiss.normalize_L2(features)

    if pred == 1:
        return 'Wajah tidak terdeteksi'
    
    faiss_index, labels, db = load_db("users/face_features.csv")

    if faiss_index is None:
        return "Database kosong"

    D, I = faiss_index.search(features, k=1) 
    score = float(D[0][0])
    idx = int(I[0][0])

    if score < 0.7:  
        return f"Tidak dikenali"
    else:
        return f"Terkenali sebagai: {labels[idx]} - (score={score:.2f})"
    
def UploadVoice(audio_file):
    if audio_file is None:
        return "❌ Tidak ada file audio. Coba rekam lagi.", "", ""

    try:
        with open(audio_file, "rb") as f:
            response = requests.post(
                "https://n8n.smartid.co.id/webhook/voice-upload",
                data=f,
                headers={"Content-Type": "audio/wav"},
            )

        if response.status_code != 200:
            return f"❌ Gagal upload. Status: {response.status_code}", "", ""
      
        try:
            data = response.json()  
        except Exception:
            return f"❌ Server tidak mengirim JSON. Balasan:\n{response}", "", ""
    
        transcription = data.get("transcribe", "(Tidak ada transkripsi)")
        summary = data.get("summary", "(Tidak ada ringkasan)")

        return "✅ Rekaman berhasil diproses", f"Transcribe: {transcription}", f"Summary: {summary}"
    except Exception as e:
        return f"❌ Gagal upload: {e}.", f"Transcribe: ", f"Summary: "