main.py

import cv2
import numpy as np
import re
from difflib import get_close_matches
from fastapi import FastAPI, UploadFile, File
from fastapi.middleware.cors import CORSMiddleware
from paddleocr import PaddleOCR

app = FastAPI(title="OCR KTP API")

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

@app.get("/")
async def root():
    return {"status": "online", "message": "KTP OCR Engine is running"}

print("Memuat model PaddleOCR...")
ocr = PaddleOCR(
    use_angle_cls=True,
    lang="en",
    det_db_thresh=0.3,
    det_db_unclip_ratio=1.8,
    rec_batch_num=6,
    use_space_char=True,
    show_log=False,
)
print("Model berhasil dimuat!")

# ============================================================
# DATABASE KOTA INDONESIA (untuk fuzzy matching nama kota di TTL)
# ============================================================
KOTA_DB = [
    "SURABAYA","JAKARTA","BANDUNG","SEMARANG","MALANG","MEDAN","MAKASSAR",
    "PALEMBANG","TANGERANG","DEPOK","BEKASI","BOGOR","YOGYAKARTA","SOLO",
    "DENPASAR","MANADO","PONTIANAK","BANJARMASIN","SAMARINDA","BALIKPAPAN",
    "PADANG","PEKANBARU","JAMBI","LAMPUNG","MATARAM","KUPANG","AMBON",
    "JAYAPURA","SORONG","TERNATE","KENDARI","PALU","GORONTALO","MAMUJU",
    "BENGKULU","BANGKA","PANGKAL PINANG","SERANG","CILEGON","TASIKMALAYA",
    "CIREBON","SUKABUMI","PURWOKERTO","TEGAL","PEKALONGAN","MAGELANG",
    "KLATEN","SURAKARTA","KEDIRI","BLITAR","PROBOLINGGO","PASURUAN",
    "MOJOKERTO","MADIUN","JEMBER","BANYUWANGI","SIDOARJO","GRESIK",
    "LAMONGAN","TUBAN","LUMAJANG","BONDOWOSO","SITUBONDO","NGAWI",
    "MAGETAN","PONOROGO","TRENGGALEK","TULUNGAGUNG","PACITAN",
    "BANGKALAN","SAMPANG","PAMEKASAN","SUMENEP","NGANJUK","JOMBANG",
    "BOJONEGORO","GARUT","SUBANG","KARAWANG","PURWAKARTA","CIANJUR",
    "BREBES","DEMAK","KUDUS","JEPARA","REMBANG","BLORA","WONOGIRI",
]

PROVINSI_CODES = {
    "11","12","13","14","15","16","17","18","19","21",
    "31","32","33","34","35","36","51","52","53","61",
    "62","63","64","65","71","72","73","74","75","76","81","82","91","94",
}


# ============================================================
# PREPROCESSING
# ============================================================
def deskew_image(img):
    """Koreksi kemiringan teks otomatis."""
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)[1]
    coords = np.column_stack(np.where(thresh > 0))
    if len(coords) < 50:
        return img
    angle = cv2.minAreaRect(coords)[-1]
    if angle < -45:
        angle = -(90 + angle)
    else:
        angle = -angle
    if abs(angle) < 0.5 or abs(angle) > 15:
        return img  # Skip jika sudut terlalu kecil atau terlalu besar (bukan skew)
    h, w = img.shape[:2]
    center = (w // 2, h // 2)
    M = cv2.getRotationMatrix2D(center, angle, 1.0)
    return cv2.warpAffine(img, M, (w, h), flags=cv2.INTER_CUBIC, borderMode=cv2.BORDER_REPLICATE)


def smart_resize(img, target_width=1280):
    """Resize ke lebar optimal untuk OCR."""
    h, w = img.shape[:2]
    if w < 800 or w > 2500:
        scale = target_width / w
        img = cv2.resize(img, (int(w * scale), int(h * scale)), interpolation=cv2.INTER_CUBIC)
    return img


def preprocess_path_a(img):
    """Path A: CLAHE + Sharpening — untuk foto dengan pencahayaan normal."""
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
    enhanced = clahe.apply(gray)
    # Unsharp mask
    blur = cv2.GaussianBlur(enhanced, (0, 0), 3.0)
    sharpened = cv2.addWeighted(enhanced, 1.5, blur, -0.5, 0)
    return cv2.cvtColor(sharpened, cv2.COLOR_GRAY2BGR)


def preprocess_path_b(img):
    """Path B: Adaptive Threshold — untuk foto gelap/kontras rendah."""
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    denoised = cv2.fastNlMeansDenoising(gray, h=10)
    thresh = cv2.adaptiveThreshold(denoised, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 21, 10)
    kernel = np.ones((1, 1), np.uint8)
    cleaned = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel)
    return cv2.cvtColor(cleaned, cv2.COLOR_GRAY2BGR)


def check_image_quality(img):
    """Cek kualitas gambar — return score 0-100."""
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    # Laplacian variance = ukuran ketajaman
    sharpness = cv2.Laplacian(gray, cv2.CV_64F).var()
    # Kontras
    contrast = gray.std()
    score = min(100, (sharpness / 100) * 50 + (contrast / 60) * 50)
    return score


def preprocess_image(image_bytes):
    """Pipeline preprocessing lengkap."""
    nparr = np.frombuffer(image_bytes, np.uint8)
    img = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
    if img is None:
        raise ValueError("Gambar tidak valid")

    img = smart_resize(img)
    img = deskew_image(img)
    return img


# ============================================================
# OCR TEXT FIXES
# ============================================================
def fix_ocr_digits(text):
    """Fix karakter yang salah di konteks ANGKA: O→0, I→1, S→5.
    Khusus NIK: tidak fix 4→9 atau 9→4 karena ambigu,
    tapi fix karakter huruf yang mirip angka."""
    return text.replace("O", "0").replace("o", "0").replace("I", "1").replace("l", "1").replace("S", "5").replace("s", "5")


def fix_nik_string(text):
    """Fix string khusus untuk NIK: O→0, I→1, S→5, B→8.
    TIDAK mengubah angka ke angka lain (4/9, 1/7) karena bisa merusak NIK asli."""
    result = []
    for c in text:
        if c in 'Oo': result.append('0')
        elif c in 'IilL': result.append('1')
        elif c in 'Ss': result.append('5')
        elif c == 'B': result.append('8')
        else: result.append(c)
    return ''.join(result)


def fix_ocr_text(text):
    """Fix karakter yang salah di konteks TEKS: 0→O, 1→I, 5→S."""
    return text.replace("0", "O").replace("1", "I").replace("5", "S")


def fix_abbreviated_address(text):
    text = re.sub(r"\bJL([A-Z])", r"JL. \1", text)
    text = re.sub(r"\bJEND\.?\s*([A-Z])", r"JEND. \1", text)
    text = re.sub(r"\bJL\s*([A-Z]{2,})", r"JL. \1", text)
    text = re.sub(r"\bJln\.?\s*", "JL. ", text)
    text = re.sub(r"\bjl\.?\s*", "JL. ", text)
    text = re.sub(r"([A-Z]{2,})NO\.?\s*(\d)", r"\1 NO. \2", text)
    text = re.sub(r"([A-Z]{2,})\.(\d)", r"\1 NO. \2", text)
    text = re.sub(r"\bNO\.?\s*(\d)", r"NO. \1", text)
    return text


def fuzzy_city(ocr_text):
    """Cocokkan nama kota OCR dengan database kota Indonesia."""
    if not ocr_text or len(ocr_text) < 3:
        return ocr_text
    matches = get_close_matches(ocr_text.upper(), KOTA_DB, n=1, cutoff=0.6)
    return matches[0] if matches else ocr_text


def validate_nik(nik):
    """Validasi NIK Indonesia: 16 digit, kode provinsi valid, tanggal valid."""
    if len(nik) != 16 or not nik.isdigit():
        return False
    if nik[:2] not in PROVINSI_CODES:
        return False
    day = int(nik[6:8])
    if not (1 <= day <= 71):
        return False
    month = int(nik[8:10])
    if not (1 <= month <= 12):
        return False
    return True


# ============================================================
# DUAL-PIPELINE OCR
# ============================================================
def run_ocr_with_confidence(img):
    """Jalankan OCR dan return lines dengan confidence score."""
    results = ocr.ocr(img, cls=True)
    if not results or not results[0]:
        return [], 0.0

    lines = []
    total_conf = 0
    count = 0
    for line in results[0]:
        text = line[1][0].strip().upper()
        conf = line[1][1]
        if conf < 0.4:  # Filter noise
            continue
        text = text.replace("?", "7").replace("!", "1")
        lines.append({"text": text, "conf": conf, "box": line[0]})
        total_conf += conf
        count += 1

    avg_conf = total_conf / count if count > 0 else 0
    return lines, avg_conf


def score_extracted_data(data, lines, avg_conf):
    score = 0
    nik = data.get("nik", "-")

    if validate_nik(nik):
        score += 45
    elif nik != "-" and len(re.sub(r"\D", "", nik)) == 16:
        score += 35

    field_weights = {
        "nama": 16,
        "ttl": 8,
        "jk": 5,
        "alamat": 10,
        "rtrw": 5,
        "keldesa": 4,
        "kecamatan": 4,
        "agama": 3,
        "pekerjaan": 3,
        "berlakuHingga": 2,
    }
    for field, weight in field_weights.items():
        value = data.get(field, "-")
        if value and value != "-":
            score += weight

    score += min(8, len(lines) * 0.6)
    score += min(8, avg_conf * 8)
    return min(100, round(score, 2))


def evaluate_ocr_candidate(lines, avg_conf, label):
    data = extract_ktp_data(lines)
    score = score_extracted_data(data, lines, avg_conf)
    return {
        "label": label,
        "lines": lines,
        "conf": avg_conf,
        "data": data,
        "score": score,
    }


def is_good_enough(candidate):
    data = candidate["data"]
    nik = data.get("nik", "-")
    if validate_nik(nik):
        return True
    if nik != "-" and len(re.sub(r"\D", "", nik)) == 16 and candidate["conf"] >= 0.35:
        return True
    return candidate["score"] >= 68


def best_candidate(candidates):
    return max(candidates, key=lambda c: (c["score"], c["conf"], len(c["lines"])))


def dual_pipeline_ocr(image_bytes):
    """Dual-pipeline: jalankan 2 preprocessing, ambil hasil terbaik."""
    nparr = np.frombuffer(image_bytes, np.uint8)
    img = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
    if img is None:
        raise ValueError("Gambar tidak valid")

    img = smart_resize(img)
    img = deskew_image(img)

    # Quality check
    quality = check_image_quality(img)
    candidates = []

    # Path A: CLAHE + Sharpening
    img_a = preprocess_path_a(img)
    lines_a, conf_a = run_ocr_with_confidence(img_a)
    candidate_a = evaluate_ocr_candidate(lines_a, conf_a, "clahe")
    candidates.append(candidate_a)

    if is_good_enough(candidate_a):
        metadata = {
            "ocr_passes": 1,
            "best_pipeline": candidate_a["label"],
            "best_score": candidate_a["score"],
            "best_confidence": round(candidate_a["conf"], 4),
        }
        return candidate_a["lines"], quality, metadata, candidate_a["data"]

    # Path B: Adaptive Threshold (fallback untuk foto jelek)
    img_b = preprocess_path_b(img)
    lines_b, conf_b = run_ocr_with_confidence(img_b)
    candidate_b = evaluate_ocr_candidate(lines_b, conf_b, "threshold")
    candidates.append(candidate_b)

    # Pilih yang terbaik, atau merge
    if conf_a >= conf_b:
        primary, secondary = lines_a, lines_b
    else:
        primary, secondary = lines_b, lines_a

    # Merge: tambahkan baris dari secondary yang tidak ada di primary
    primary_texts = {l["text"] for l in primary}
    for line in secondary:
        if line["text"] not in primary_texts and line["conf"] > 0.6:
            primary.append(line)

    merged_conf = max(conf_a, conf_b)
    candidates.append(evaluate_ocr_candidate(primary, merged_conf, "merged"))

    current_best = best_candidate(candidates)
    if current_best["data"].get("nik", "-") == "-":
        h, w = img.shape[:2]
        top_region = img[:max(1, int(h * 0.45)), 0:w]
        top_a = preprocess_path_a(top_region)
        lines_top, conf_top = run_ocr_with_confidence(top_a)
        top_candidate = evaluate_ocr_candidate(lines_top, conf_top, "nik_region")
        candidates.append(top_candidate)
        if top_candidate["data"].get("nik", "-") != "-":
            combined_lines = list(current_best["lines"])
            combined_texts = {l["text"] for l in combined_lines}
            for line in lines_top:
                if line["text"] not in combined_texts:
                    combined_lines.append(line)
            candidates.append(evaluate_ocr_candidate(combined_lines, max(current_best["conf"], conf_top), "merged_nik_region"))
        current_best = best_candidate(candidates)

    if not is_good_enough(current_best) and current_best["score"] < 45:
        for label, rotate_code in [
            ("rotate_90_clockwise", cv2.ROTATE_90_CLOCKWISE),
            ("rotate_90_counterclockwise", cv2.ROTATE_90_COUNTERCLOCKWISE),
        ]:
            rotated = cv2.rotate(img, rotate_code)
            rotated_a = preprocess_path_a(rotated)
            lines_r, conf_r = run_ocr_with_confidence(rotated_a)
            candidates.append(evaluate_ocr_candidate(lines_r, conf_r, label))

    current_best = best_candidate(candidates)
    metadata = {
        "ocr_passes": len(candidates),
        "best_pipeline": current_best["label"],
        "best_score": current_best["score"],
        "best_confidence": round(current_best["conf"], 4),
    }
    return current_best["lines"], quality, metadata, current_best["data"]


# ============================================================
# SPATIAL HELPERS
# ============================================================
def extract_value_after_colon(text):
    """Ambil teks setelah ':' atau ';'. Jika tidak ada, kembalikan asli."""
    match = re.search(r'[:;]\s*(.+)', text)
    return match.group(1).strip() if match else text.strip()


def sort_lines_by_y(ocr_lines):
    """Urutkan baris OCR berdasarkan posisi Y (atas ke bawah)."""
    def get_y(line):
        box = line["box"]
        return (box[0][1] + box[2][1]) / 2
    return sorted(ocr_lines, key=get_y)


def merge_same_y_lines(sorted_lines, y_threshold=15):
    """Gabungkan baris yang posisi Y-nya berdekatan (satu baris visual)."""
    if not sorted_lines:
        return []
    merged = []
    group = [sorted_lines[0]]
    for i in range(1, len(sorted_lines)):
        prev_y = (group[-1]["box"][0][1] + group[-1]["box"][2][1]) / 2
        curr_y = (sorted_lines[i]["box"][0][1] + sorted_lines[i]["box"][2][1]) / 2
        if abs(curr_y - prev_y) < y_threshold:
            group.append(sorted_lines[i])
        else:
            group.sort(key=lambda l: (l["box"][0][0] + l["box"][1][0]) / 2)
            text = " ".join([l["text"] for l in group])
            conf = sum(l["conf"] for l in group) / len(group)
            merged.append({"text": text, "conf": conf, "box": group[0]["box"]})
            group = [sorted_lines[i]]
    if group:
        group.sort(key=lambda l: (l["box"][0][0] + l["box"][1][0]) / 2)
        text = " ".join([l["text"] for l in group])
        conf = sum(l["conf"] for l in group) / len(group)
        merged.append({"text": text, "conf": conf, "box": group[0]["box"]})
    return merged


# ============================================================
# DATA EXTRACTION (LABEL-PATTERN REGEX — no spatial ordering)
# ============================================================
def fix_date_string(raw):
    """Perbaiki karakter non-angka dalam string tanggal: O→0, I→1, S→5, l→1."""
    return re.sub(r"[OoIlSs]", lambda m: {"O":"0","o":"0","I":"1","l":"1","S":"5","s":"5"}[m.group()], raw)


def extract_ktp_data(ocr_lines):
    data = {
        "nik": "-", "nama": "-", "ttl": "-", "jk": "-", "goldarah": "-",
        "alamat": "-", "rtrw": "-", "keldesa": "-", "kecamatan": "-",
        "agama": "-", "status": "-", "pekerjaan": "-",
        "kewarganegaraan": "-", "berlakuHingga": "-",
    }
    if not ocr_lines:
        return data

    sorted_lines = sort_lines_by_y(ocr_lines)
    merged = merge_same_y_lines(sorted_lines)
    raw_lines = [l["text"] for l in merged]
    # Gabungkan semua teks menjadi satu string bersih untuk regex global
    full_text = " ".join(raw_lines)
    # Versi full_text dengan fix teks (0→O, 1→I) untuk field berbasis teks
    full_text_fixed = fix_ocr_text(full_text)

    # ── Helper ──────────────────────────────────────────────────────────────
    def search(patterns, text=full_text, flags=re.IGNORECASE):
        """Coba daftar regex berurutan, return group(1) pertama yang match."""
        for p in patterns:
            m = re.search(p, text, flags)
            if m:
                return m.group(1).strip()
        return None

    # ── NIK ─────────────────────────────────────────────────────────────────
    # Gunakan fix_nik_string: hanya fix huruf→angka, tidak ubah angka→angka
    nik_raw = re.sub(r"\s", "", full_text)
    nik_fixed = fix_nik_string(nik_raw)
    m = re.search(r"(\d{16})", nik_fixed)
    if m:
        data["nik"] = m.group(1)

    # ── NAMA ─────────────────────────────────────────────────────────────────
    # Cari dengan label dulu (lebih akurat), lalu fallback spatial (baris setelah NIK)
    nama_val = search([
        r"(?:NAMA|NARNA|N[A4]M[A4]|LAMA|L[A4]M[A4])\s*[:.]*\s*([A-Z][A-Z\s.,'-]{2,49}?)\s*(?=TEMPAT|TGL|LAHIR|JENIS|KELAMIN|LAKI|PEREMPUAN|GOL|DARAH|ALAMA?T|BERLAKU|$)",
        # Fallback: cari pola 'Nama : XXX' secara lebih longgar
        r"N[A4A]M[A4A]\s*[:.]+\s*([A-Z][A-Z\s.'-]{2,49})",
    ], full_text_fixed)
    if nama_val:
        nama_val = re.sub(r"[^A-Z\s.'-]", "", nama_val).strip()
        if 2 < len(nama_val) <= 50:
            data["nama"] = nama_val
    # Fallback spasial: baris pertama setelah NIK yang hanya berisi huruf
    if data["nama"] == "-":
        nik_idx = next((i for i, l in enumerate(raw_lines) if re.search(r"\d{14,16}", re.sub(r"\s","",l))), -1)
        if nik_idx != -1:
            for candidate in raw_lines[nik_idx+1:nik_idx+3]:
                c = re.sub(r"(?i)(nama|lama|n[a4]m[a4])[\s:.]?", "", candidate).strip()
                c = re.sub(r"[^A-Z\s.'-]", "", c.upper()).strip()
                if 3 < len(c) <= 50 and not any(kw in c for kw in ["LAHIR","JENIS","LAKI","ALAMA","RT","RW","JL","AGAMA","KAWIN"]):
                    data["nama"] = c
                    break

    # ── TTL ──────────────────────────────────────────────────────────────────
    # Tangkap tanggal yang toleran terhadap O/I/S, lalu fix
    ttl_pat = r"(?:TEMPAT[/\s]*TGL[/\s]*LAHIR|TGL\s*LAHIR|LAHIR|TMPTTL|T\.?T\.?L\.?)\s*[:.]*\s*(.*?)(?=JENIS|KELAMIN|LAKI|PEREMPUAN|GOL|DARAH|ALAMA?T|$)"
    m = re.search(ttl_pat, full_text, re.IGNORECASE)
    if m:
        raw_ttl = m.group(1).strip()
        # Cari tanggal toleran: digit atau O,I,S,l
        date_m = re.search(r"([0-9OIlS]{2}[-./]?[0-9OIlS]{2}[-./]?[0-9OIlS]{4})", raw_ttl, re.IGNORECASE)
        if date_m:
            clean_date = fix_date_string(date_m.group(1))
            clean_date = re.sub(r"[./]", "-", clean_date)
            # Kota: teks alfabet sebelum tanggal
            city_part = raw_ttl[:date_m.start()].strip()
            city_part = re.sub(r"[^A-Z\s]", "", fix_ocr_text(city_part)).strip()
            city = fuzzy_city(city_part) if city_part else ""
            data["ttl"] = f"{city}, {clean_date}" if city else clean_date
    # Fallback: cari tanggal di mana saja
    if data["ttl"] == "-":
        date_m = re.search(r"([0-9OIlS]{2}[-][0-9OIlS]{2}[-][0-9OIlS]{4})", full_text, re.IGNORECASE)
        if date_m and "BERLAKU" not in full_text[max(0, date_m.start()-10):date_m.start()]:
            data["ttl"] = fix_date_string(date_m.group(1))

    # ── JENIS KELAMIN & GOL DARAH ────────────────────────────────────────────
    for line in raw_lines:
        uline = line.upper()
        if ("PEREMPUAN" in uline or "LAKI" in uline) and data["jk"] == "-":
            data["jk"] = "PEREMPUAN" if "PEREMPUAN" in uline else "LAKI-LAKI"
        # Gol darah: cari di dekat konteks GOL/DARAH, "0" → "O"
        if data["goldarah"] == "-":
            gd = re.search(r"(?:GOL|DARAH)[^A-Z0]*([ABO0]{1,2})\b", uline)
            if gd:
                val = gd.group(1).replace("0", "O")
                if val in ["A", "B", "AB", "O"]:
                    data["goldarah"] = val
    # Fallback Gol Darah: dari akhir baris JK
    if data["goldarah"] == "-":
        for line in raw_lines:
            parts = line.upper().split()
            if parts and parts[-1].replace("0","O") in ["A","B","AB","O"]:
                if "LAKI" in line or "PEREMPUAN" in line:
                    data["goldarah"] = parts[-1].replace("0","O")
                    break

    # ── ALAMAT ───────────────────────────────────────────────────────────────
    # Multi-barrier lookahead + max 100 karakter
    alamat_val = search([
        r"(?:ALAMA?T|ALAM[A4]T)\s*[:.]*\s*([A-Z0-9\s.,'/\-]{5,100}?)\s*(?=R[TW7][\s/]|RT[\s/]|RW[\s/]|KEL|DESA|KEC|AGAMA|STATUS|PEKERJAAN|$)",
    ])
    if alamat_val:
        data["alamat"] = fix_abbreviated_address(alamat_val.strip())
    # Fallback: cari baris yang punya keyword jalan
    if data["alamat"] == "-":
        for line in raw_lines:
            if re.search(r"\b(JL\.?|JLN\.?|GANG|GG\.?|JEND)\b", line, re.IGNORECASE):
                data["alamat"] = fix_abbreviated_address(re.sub(r"(?i)^alama?t\s*[:.]*\s*", "", line).strip())
                break

    # ── RT/RW ────────────────────────────────────────────────────────────────
    m = re.search(r"(\d{1,3})\s*[/|\\]\s*(\d{1,3})", full_text)
    if m:
        data["rtrw"] = f"{m.group(1).zfill(3)}/{m.group(2).zfill(3)}"

    # ── KELURAHAN/DESA ────────────────────────────────────────────────────────
    keldesa_val = search([
        r"(?:KEL\.?/?DESA|KELURAHAN|DESA)\s*[:.]*\s*([A-Z][A-Z\s.-]{2,49}?)\s*(?=KEC|KABUPATEN|KOTA|PROVINSI|AGAMA|$)",
    ], full_text_fixed)
    if keldesa_val:
        keldesa_val = re.sub(r"[^A-Z\s.-]", "", keldesa_val).strip()
        if len(keldesa_val) > 2:
            data["keldesa"] = keldesa_val

    # ── KECAMATAN ─────────────────────────────────────────────────────────────
    kec_val = search([
        r"(?:KECAMATAN|KEC\.?)\s*[:.]*\s*([A-Z][A-Z\s.-]{2,49}?)\s*(?=KABUPATEN|KOTA|PROVINSI|AGAMA|STATUS|$)",
    ], full_text_fixed)
    if kec_val:
        kec_val = re.sub(r"[^A-Z\s.-]", "", kec_val).strip()
        if len(kec_val) > 2:
            data["kecamatan"] = kec_val

    # ── AGAMA ─────────────────────────────────────────────────────────────────
    # Fuzzy: toleran terhadap J→I, 4→A, dll.
    AGAMA_LIST = ["ISLAM","KRISTEN","PROTESTAN","KATHOLIK","KATOLIK","HINDU","BUDHA","BUDDHA","KONGHUCU"]
    for ag in AGAMA_LIST:
        # Izinkan 1 karakter berbeda (misalnya J vs I di JSLAM)
        if ag in full_text_fixed:
            data["agama"] = ag
            break
    if data["agama"] == "-":
        # Fuzzy: cari sub-string yang mirip dengan toleransi 1 karakter
        for ag in AGAMA_LIST:
            pattern = "".join(f"[{c}{c.lower()}{'J' if c=='I' else ''}{'4' if c=='A' else ''}{'1' if c=='I' else ''}]" for c in ag)
            if re.search(pattern, full_text_fixed, re.IGNORECASE):
                data["agama"] = ag
                break

    # ── STATUS ────────────────────────────────────────────────────────────────
    sm = re.search(r"\b(BELUM\s*KAWIN|KAWIN|CERAI\s*HIDUP|CERAI\s*MATI)\b", full_text)
    if sm:
        v = sm.group(0)
        data["status"] = "BELUM KAWIN" if "BELUM" in v else re.sub(r"\s+", " ", v)

    # ── PEKERJAAN ─────────────────────────────────────────────────────────────
    JOBS = [
        "PELAJAR/MAHASISWA","PELAJAR","MAHASISWA","WIRASWASTA",
        "KARYAWAN SWASTA","KARYAWAN","PEGAWAI NEGERI SIPIL","PEGAWAI NEGERI",
        "BURUH","MENGURUS RUMAH TANGGA","BELUM/TIDAK BEKERJA","GURU","DOSEN",
        "PNS","TNI","POLRI","PEDAGANG","PETANI","NELAYAN","DOKTER","PERAWAT",
    ]
    for line in raw_lines:
        for job in JOBS:
            if job in line.upper():
                data["pekerjaan"] = job
                break
        if data["pekerjaan"] != "-":
            break

    # ── KEWARGANEGARAAN ───────────────────────────────────────────────────────
    km = re.search(r"\b(WNI|WN\s*I|W\s*N\s*I|WN1|WNA)\b", full_text)
    if km:
        data["kewarganegaraan"] = "WNA" if "A" in km.group(0) else "WNI"

    # ── BERLAKU HINGGA ────────────────────────────────────────────────────────
    if re.search(r"\bSEUMUR\s*HIDUP\b", full_text):
        data["berlakuHingga"] = "SEUMUR HIDUP"
    else:
        bh = re.search(r"BERLAKU\s*HINGGA\s*[:.]*\s*([0-9OIlS]{2}[-][0-9OIlS]{2}[-][0-9OIlS]{4})", full_text, re.IGNORECASE)
        if bh:
            data["berlakuHingga"] = fix_date_string(bh.group(1))

    # ── CLEANUP ───────────────────────────────────────────────────────────────
    for k in data:
        if data[k] and data[k] != "-":
            data[k] = re.sub(r"^[:\-\s.,]+|[:\-\s.,]+$", "", data[k]).strip()
            if not data[k]:
                data[k] = "-"

    return data



# ============================================================
# API ENDPOINT
# ============================================================
@app.post("/scan-ktp")
async def scan_ktp(file: UploadFile = File(...)):
    try:
        image_bytes = await file.read()
        ocr_lines, quality_score, metadata, extracted_data = dual_pipeline_ocr(image_bytes)

        if extracted_data["nik"] == "-" and extracted_data["nama"] == "-":
            return {
                "status": "error",
                "message": "NIK belum terbaca. Coba foto ulang dengan area NIK lebih dekat, terang, dan tidak terpotong.",
                "quality_score": round(quality_score, 2),
                "ocr_metadata": metadata,
            }

        warning = None
        if extracted_data["nik"] == "-":
            warning = "NIK belum terbaca, data perlu dicek manual sebelum disimpan."
        elif quality_score < 30 or metadata.get("best_score", 0) < 55:
            warning = "Kualitas foto kurang baik, beberapa data mungkin tidak akurat."

        result = {
            "status": "success",
            "data": extracted_data,
            "quality_score": round(quality_score, 2),
            "ocr_metadata": metadata,
        }
        if warning:
            result["warning"] = warning
        return result

    except ValueError as e:
        return {"status": "error", "message": str(e)}
    except Exception as e:
        import traceback
        return {
            "status": "error",
            "message": f"Terjadi kesalahan internal. Detail: {str(e)}",
            "traceback": traceback.format_exc(),
        }


if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)