kaderkita_ocr 0.3.0

On-device Flutter OCR for Indonesian KTP and Kartu Keluarga (KK). Camera scanners with optional debug pipeline / ROI inspection.

kaderkita_ocr

On-device Flutter OCR for Indonesian KTP and Kartu Keluarga (KK).

Scanners capture from the camera, run Google ML Kit Text Recognition locally, and return structured fields. Optional debug mode attaches raw OCR text, confidence, ROI/crop previews, and step-by-step pipeline images for tooling.

Privacy: all OCR runs on-device. No images or text are uploaded by this package.

Requirements

Flutter	>=3.38.0
Dart	>=3.10.0
Platforms	Android, iOS

Native pieces come from transitive plugins: camera and google_mlkit_text_recognition.

Install

dependencies:
  kaderkita_ocr: ^0.3.0

Platform setup

Android — camera permission in AndroidManifest.xml:

<uses-permission android:name="android.permission.CAMERA"/>

iOS — camera usage string in Info.plist:

<key>NSCameraUsageDescription</key>
<string>Camera is used to scan KTP and Kartu Keluarga on-device.</string>

Quick start

import 'package:kaderkita_ocr/kaderkita_ocr.dart';

Future<void> scanKtp(BuildContext context) async {
  final result = await Navigator.push<KtpScanResult>(
    context,
    MaterialPageRoute(
      builder: (_) => const KtpScanner(
        // debug: true → attach ScanDebug (raw text, ROIs, pipeline images)
        config: KaderkitaOcrConfig(debug: false),
      ),
    ),
  );
  if (result == null) return;
  print(result.data.nik);
  print(result.data.nama);
}

Same pattern for KK:

final result = await Navigator.push<KkScanResult>(
  context,
  MaterialPageRoute(builder: (_) => const KkScanner()),
);

Public API

Type	Role
KtpScanner / KkScanner	Full-screen camera scanners; pop with a result
KaderkitaOcrConfig	debug, autoCapture
KtpScanResult / KkScanResult	data + optional debug
KtpData / KkData / KkMember	Structured fields
ScanDebug / RoiPreview / ImagePipelineStep	Debug payload
ScanDebugPanel / ImagePipelineViewer	Optional debug UI widgets
NikValidator	Structural NIK checks

Internals (OcrService, parsers, Monte Carlo tools) live under lib/src/ and are not part of the stable public API.

Data model / output

Each scanner pops a result whose data holds the structured fields (always present) and an optional debug payload (only when config.debug == true). Every field is nullable — OCR may not recover it.

KtpData (KtpScanResult.data)

Field	Type	toJson key
nik	String?	nik
nama	String?	nama
tempatLahir	String?	tempat_lahir
tanggalLahir	String?	tanggal_lahir (DD-MM-YYYY)
jenisKelamin	String?	jenis_kelamin
confidence	double?	confidence
rawText	String?	— (in-memory only)

Helper: isComplete (has nik, nama, and tanggalLahir).

KkData (KkScanResult.data)

Field	Type	toJson key
nomorKk	String?	nomor_kk
namaKepalaKeluarga	String?	nama_kepala_keluarga
alamat	String?	alamat
rtRw	String?	rt_rw
kodePos	String?	kode_pos
desaKelurahan	String?	desa_kelurahan
kecamatan	String?	kecamatan
kabupatenKota	String?	kabupaten_kota
provinsi	String?	provinsi
members	List<KkMember>	members (each via toFieldMap())
confidence	double?	confidence
rawText	String?	— (in-memory only)

Helper: hasMembers.

KkMember (a row of the family table)

All fields are String? except the flag: no, namaLengkap, nik, jenisKelamin, tempatLahir, tanggalLahir, agama, pendidikan, jenisPekerjaan, golonganDarah, and isKepalaKeluarga (bool).

isKepalaKeluarga marks the head of family — the member whose name matches the extracted namaKepalaKeluarga (exact, else nearest by Levenshtein). toFieldMap() returns Map<String, dynamic> with snake_case keys (nama_lengkap, nik, is_kepala_keluarga, …). Helpers: hasIdentity, isEmptyRow.

golonganDarah is retained in the model and output (golongan_darah) but is not scanned by default, so it is typically null.

Debug mode

const KaderkitaOcrConfig(debug: true)

When enabled, result.debug includes:

• rawText — OCR text dump
• confidence — aggregate score when available
• rois — bracket / column crop previews
• pipelineSteps — labeled images for each preprocess / layout stage

Show them with ScanDebugPanel or ImagePipelineViewer.

When debug is false (default), heavy image bytes are not attached to the result.

Documentation

• KTP NIK glyph-compare pipeline — Methods A/B/C preprocessing, crop geometry, multi-example templates, voting, and all tunable parameters.
• Nomor KK glyph-compare pipeline — how the KK number reuses the same color pipeline (layout differences + final-capture gating).

Example

See example/ for a minimal host app with a debug toggle and result pages.

cd example && flutter run

License

Apache License 2.0 — see LICENSE.