blitzid 0.3.1__tar.gz → 0.3.2__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {blitzid-0.3.1 → blitzid-0.3.2}/.github/workflows/ci.yml +3 -1
- {blitzid-0.3.1 → blitzid-0.3.2}/CHANGELOG.md +14 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/Dockerfile +5 -4
- {blitzid-0.3.1 → blitzid-0.3.2}/PKG-INFO +86 -14
- {blitzid-0.3.1 → blitzid-0.3.2}/README.md +82 -13
- {blitzid-0.3.1 → blitzid-0.3.2}/docs/coding-standard.md +20 -10
- {blitzid-0.3.1 → blitzid-0.3.2}/docs/integration.md +17 -12
- {blitzid-0.3.1 → blitzid-0.3.2}/pyproject.toml +8 -2
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/__init__.py +9 -1
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/_dates.py +8 -2
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/api/_analyze.py +28 -3
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/api/_app.py +42 -18
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/exceptions.py +4 -0
- blitzid-0.3.2/src/blitzid/reading/barcode.py +364 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_api.py +82 -1
- blitzid-0.3.2/tests/test_barcode.py +396 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/uv.lock +714 -792
- {blitzid-0.3.1 → blitzid-0.3.2}/.dockerignore +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/.env.example +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/.github/workflows/docker.yml +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/.github/workflows/publish.yml +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/.gitignore +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/.pre-commit-config.yaml +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/CONTRIBUTING.md +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/LICENSE +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/docker-compose.dev.yml +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/docker-compose.gpu.yml +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/docker-compose.yml +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/images/README.md +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/images/bub_der_personalausweis_kopie.jpg +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/images/nl_td1_id_specimen.jpg +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/images/ny_license_plate_wrap.jpg +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/images/nys_driver_license_sample.webp +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/images/solvay_conference_1927.jpg +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/images/td3_passport_specimen.jpg +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/images/tlc_driver_license_sample.png +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/models/.gitkeep +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/models/README.md +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/scripts/README.md +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/scripts/api_smoke.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/scripts/benchmark.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/scripts/demo_cli.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/scripts/download_models.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/scripts/face_detector_demo.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/scripts/midv500.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/scripts/ocr_demo.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/scripts/structurize_smoke.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/scripts/verify_smoke.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/_image.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/_models.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/api/__init__.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/api/_jobs.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/api/_liveness.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/api/_schemas.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/api/_upload.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/face/__init__.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/face/_antispoof.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/face/_arcface.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/face/_attributes.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/face/_face.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/face/_motion.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/face/_scrfd.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/face/detector.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/face/verifier.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/py.typed +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/reading/__init__.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/reading/_observability.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/reading/consistency.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/reading/document.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/reading/mrz.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/reading/ocr.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/src/blitzid/reading/structurize.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/conftest.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_antispoof.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_attributes.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_consistency.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_detector.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_detector_helpers.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_document.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_exceptions.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_image.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_models.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_motion.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_mrz.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_ocr.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_refactored.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_structurize.py +0 -0
- {blitzid-0.3.1 → blitzid-0.3.2}/tests/test_verifier.py +0 -0
|
@@ -27,5 +27,7 @@ jobs:
|
|
|
27
27
|
- uses: astral-sh/setup-uv@v8.1.0
|
|
28
28
|
with:
|
|
29
29
|
python-version: ${{ matrix.python-version }}
|
|
30
|
-
|
|
30
|
+
# The barcode extra enables the zxing-cpp round-trip tier;
|
|
31
|
+
# without it those tests skip, not fail.
|
|
32
|
+
- run: uv sync --extra dev --extra barcode
|
|
31
33
|
- run: uv run pytest tests/ -v
|
|
@@ -4,6 +4,20 @@ All notable changes to this project will be documented in this file.
|
|
|
4
4
|
|
|
5
5
|
## [Unreleased]
|
|
6
6
|
|
|
7
|
+
## [0.3.2] - 2026-10-06
|
|
8
|
+
|
|
9
|
+
### Added
|
|
10
|
+
|
|
11
|
+
- PDF417 barcode reading for US and Canadian driver's licenses
|
|
12
|
+
(`blitzid[barcode]` extra, zxing-cpp): `BarcodeReader` /
|
|
13
|
+
`BarcodeRecord` / `BarcodeError` decode the back-of-card PDF417 and
|
|
14
|
+
parse the AAMVA DL/ID payload (versions 02–14). The parser validates
|
|
15
|
+
the header's subfile offsets, lengths, and number-of-entries count,
|
|
16
|
+
gates the mandatory fields with per-edition tag maps pinned by
|
|
17
|
+
fixture tests, and keeps every un-mapped tag raw in `extra_tags`.
|
|
18
|
+
The HTTP API gains a `barcode` analysis type and a `/health` flag;
|
|
19
|
+
a missing `barcode` extra degrades to the per-engine `400`.
|
|
20
|
+
|
|
7
21
|
## [0.3.1] - 2026-10-06
|
|
8
22
|
|
|
9
23
|
### Fixed
|
|
@@ -6,10 +6,11 @@ COPY README.md pyproject.toml uv.lock ./
|
|
|
6
6
|
COPY src/ src/
|
|
7
7
|
COPY scripts/ scripts/
|
|
8
8
|
# Every runtime feature ships in the image: the HTTP API (api), face
|
|
9
|
-
# detection/verification in the core, OCR, MRZ,
|
|
10
|
-
# extraction and Langfuse tracing
|
|
11
|
-
# (scripts). Only the dev toolchain
|
|
12
|
-
|
|
9
|
+
# detection/verification in the core, OCR, MRZ, PDF417 barcode reading
|
|
10
|
+
# (barcode), LLM-backed structured extraction and Langfuse tracing
|
|
11
|
+
# (ocr), and the bundled scripts (scripts). Only the dev toolchain
|
|
12
|
+
# stays out.
|
|
13
|
+
RUN uv sync --frozen --no-dev --extra api --extra barcode --extra ocr --extra scripts
|
|
13
14
|
# Bake every engine's weights (SCRFD, ArcFace, RapidOCR) so cold start
|
|
14
15
|
# needs no downloads.
|
|
15
16
|
RUN uv run --no-sync python scripts/download_models.py --models-dir /models
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.5
|
|
2
2
|
Name: blitzid
|
|
3
|
-
Version: 0.3.
|
|
3
|
+
Version: 0.3.2
|
|
4
4
|
Summary: Modular DNN-based ID document reading framework (face detection, OCR, MRZ) optimized for CPU deployment
|
|
5
5
|
Project-URL: Homepage, https://github.com/malvavisc0/blitzid
|
|
6
6
|
Project-URL: Repository, https://github.com/malvavisc0/blitzid
|
|
@@ -38,11 +38,14 @@ Requires-Dist: ruff; extra == 'all'
|
|
|
38
38
|
Requires-Dist: tqdm; extra == 'all'
|
|
39
39
|
Requires-Dist: types-tqdm; extra == 'all'
|
|
40
40
|
Requires-Dist: uvicorn[standard]; extra == 'all'
|
|
41
|
+
Requires-Dist: zxing-cpp; extra == 'all'
|
|
41
42
|
Provides-Extra: api
|
|
42
43
|
Requires-Dist: fastapi; extra == 'api'
|
|
43
44
|
Requires-Dist: python-multipart; extra == 'api'
|
|
44
45
|
Requires-Dist: redis; extra == 'api'
|
|
45
46
|
Requires-Dist: uvicorn[standard]; extra == 'api'
|
|
47
|
+
Provides-Extra: barcode
|
|
48
|
+
Requires-Dist: zxing-cpp; extra == 'barcode'
|
|
46
49
|
Provides-Extra: dev
|
|
47
50
|
Requires-Dist: fakeredis; extra == 'dev'
|
|
48
51
|
Requires-Dist: fastapi; extra == 'dev'
|
|
@@ -72,8 +75,9 @@ Description-Content-Type: text/markdown
|
|
|
72
75
|
[](LICENSE)
|
|
73
76
|
|
|
74
77
|
Read identity documents from images: find and compare faces, read printed
|
|
75
|
-
text,
|
|
76
|
-
|
|
78
|
+
text, parse the machine-readable zone, and decode the PDF417 barcode on
|
|
79
|
+
US and Canadian driver's licenses, in one typed Python library that
|
|
80
|
+
runs on an ordinary CPU.
|
|
77
81
|
|
|
78
82
|
No GPU and no CUDA stack: inference runs on onnxruntime's CPU provider,
|
|
79
83
|
face detection takes about 4 ms per image on a laptop, and the weights
|
|
@@ -85,9 +89,10 @@ results in RAM only.
|
|
|
85
89
|
|
|
86
90
|
## Why blitzid
|
|
87
91
|
|
|
88
|
-
- **One install, one API.** Face detection, OCR, MRZ parsing,
|
|
89
|
-
document crop/QC behind a single typed surface,
|
|
90
|
-
a detector, an OCR engine, and your own MRZ code
|
|
92
|
+
- **One install, one API.** Face detection, OCR, MRZ parsing, PDF417
|
|
93
|
+
barcode reading, and document crop/QC behind a single typed surface,
|
|
94
|
+
instead of stitching a detector, an OCR engine, and your own MRZ code
|
|
95
|
+
together.
|
|
91
96
|
- **Validation over guessing.** Every MRZ field is gated by ICAO 9303
|
|
92
97
|
check digits. Obvious OCR confusions (0/O, 1/I, …) are undone by
|
|
93
98
|
each field's alphabet before validation, and a zone that still does
|
|
@@ -110,6 +115,7 @@ results in RAM only.
|
|
|
110
115
|
```bash
|
|
111
116
|
pip install blitzid # or: uv add blitzid
|
|
112
117
|
pip install blitzid[ocr] # with OCR text reading (RapidOCR)
|
|
118
|
+
pip install blitzid[barcode] # with PDF417/AAMVA barcode reading (zxing-cpp)
|
|
113
119
|
pip install blitzid[api] # with the HTTP API server
|
|
114
120
|
```
|
|
115
121
|
|
|
@@ -199,6 +205,15 @@ nationality), and parses TD1 (3x30), TD2 (2x36), and TD3 (2x44) into an
|
|
|
199
205
|
convention (overflow at the start of `optional_data1`, no printed check
|
|
200
206
|
digit).
|
|
201
207
|
|
|
208
|
+
**PDF417 barcode reading.** `BarcodeReader` (the `barcode` extra,
|
|
209
|
+
zxing-cpp) decodes the PDF417 code on the back of US and Canadian
|
|
210
|
+
driver's licenses and parses the AAMVA DL/ID payload into a typed
|
|
211
|
+
`BarcodeRecord` — jurisdiction, AAMVA version, names, dates, sex, and
|
|
212
|
+
license number, plus the raw jurisdiction-specific extras. The header's
|
|
213
|
+
subfile offsets, lengths, and number-of-entries count are validated and
|
|
214
|
+
the mandatory fields are gated, so a malformed payload raises
|
|
215
|
+
`BarcodeError` naming the offending field.
|
|
216
|
+
|
|
202
217
|
**Structured extraction.** `StructuredOCRReader` (same extra) feeds the
|
|
203
218
|
OCR lines to an OpenAI-compatible chat endpoint (e.g. a local vLLM
|
|
204
219
|
server) and returns a typed `StructuredOCR` record; fields the model
|
|
@@ -221,8 +236,8 @@ face). Returns the crop plus a `QualityReport` verdict.
|
|
|
221
236
|
|
|
222
237
|
**HTTP API and Docker.** The `api` extra adds a FastAPI service: an
|
|
223
238
|
async job queue (Redis, persistence disabled) for
|
|
224
|
-
face/attributes/antispoof/OCR/MRZ/structured/consistency,
|
|
225
|
-
`/crop` (document QC), `/verify` (1:1 face comparison), and
|
|
239
|
+
face/attributes/antispoof/OCR/MRZ/barcode/structured/consistency,
|
|
240
|
+
synchronous `/crop` (document QC), `/verify` (1:1 face comparison), and
|
|
226
241
|
`/liveness` (challenge-response liveness), and `/health` for
|
|
227
242
|
orchestration.
|
|
228
243
|
`docker compose up` runs it with all weights baked in.
|
|
@@ -249,9 +264,10 @@ curl http://localhost:8000/api/health
|
|
|
249
264
|
file (any OpenCV-decodable format; PDFs get a precise `415`) and
|
|
250
265
|
`types` (repeated and/or comma-separated: `face`, `attributes`,
|
|
251
266
|
`antispoof`, `ocr`,
|
|
252
|
-
`mrz`, `structured`, `consistency` — `attributes` predicts
|
|
253
|
-
gender, and race per face, `antispoof` scores live vs
|
|
254
|
-
screen photos, `
|
|
267
|
+
`mrz`, `barcode`, `structured`, `consistency` — `attributes` predicts
|
|
268
|
+
age group, gender, and race per face, `antispoof` scores live vs
|
|
269
|
+
printed vs screen photos, `barcode` decodes the PDF417/AAMVA payload
|
|
270
|
+
from a driver's-license back, `structured` rebuilds the OCR text into a
|
|
255
271
|
typed record via the LLM, and `consistency` cross-checks that printed
|
|
256
272
|
record against the machine-readable zone and the portrait's age band
|
|
257
273
|
and gender against the birth date and sex; the photo rows need the
|
|
@@ -259,8 +275,8 @@ and gender against the birth date and sex; the photo rows need the
|
|
|
259
275
|
`BLITZID_LLM_*` config).
|
|
260
276
|
Validation is eager, before the job exists: `422` unknown type or
|
|
261
277
|
missing fields, `400` undecodable bytes or an analysis whose engine is
|
|
262
|
-
unavailable (`ocr`/`mrz` without the `ocr` extra, `
|
|
263
|
-
`consistency` without
|
|
278
|
+
unavailable (`ocr`/`mrz` without the `ocr` extra, `barcode` without the
|
|
279
|
+
`barcode` extra, `structured`/`consistency` without
|
|
264
280
|
the extra or the LLM config), `413` above the
|
|
265
281
|
upload cap, `503` + `Retry-After` when the queue is full or Redis is
|
|
266
282
|
down. Accepted jobs return `202`:
|
|
@@ -273,7 +289,7 @@ curl -F image=@id.jpg -F types=face,ocr localhost:8000/api/analyze
|
|
|
273
289
|
**GET /jobs/{job_id}** polls: `{"status": "queued"}` / `running`; done
|
|
274
290
|
returns `200` with one section per requested type (`face`,
|
|
275
291
|
`attributes`, `antispoof`, `ocr`,
|
|
276
|
-
`mrz`, `structured`, `consistency`; a failing section carries
|
|
292
|
+
`mrz`, `barcode`, `structured`, `consistency`; a failing section carries
|
|
277
293
|
`{"error": ...}` while
|
|
278
294
|
the others still return). The read claims the result, so every later
|
|
279
295
|
read gets `410 Gone`; unknown ids get `404` (an expired job reads `410` until
|
|
@@ -349,6 +365,30 @@ What a done result looks like on the specimen ID (abbreviated):
|
|
|
349
365
|
}
|
|
350
366
|
```
|
|
351
367
|
|
|
368
|
+
On a US driver's-license back, a `barcode` section looks like this:
|
|
369
|
+
|
|
370
|
+
```json
|
|
371
|
+
{
|
|
372
|
+
"record": {
|
|
373
|
+
"jurisdiction": "636014",
|
|
374
|
+
"aamva_version": "08",
|
|
375
|
+
"jurisdiction_version": "00",
|
|
376
|
+
"family_name": "PUBLIC",
|
|
377
|
+
"given_name": "JOHN",
|
|
378
|
+
"middle_name": "QUINCY",
|
|
379
|
+
"birth_date": "1989-01-24",
|
|
380
|
+
"issue_date": "2013-06-04",
|
|
381
|
+
"expiry_date": "2035-01-31",
|
|
382
|
+
"sex": "M",
|
|
383
|
+
"license_number": "D12345678",
|
|
384
|
+
"height": "069 in",
|
|
385
|
+
"extra_tags": {"ZCB": "…"},
|
|
386
|
+
"symbology": "PDF417"
|
|
387
|
+
},
|
|
388
|
+
"processing_time_ms": 87.2
|
|
389
|
+
}
|
|
390
|
+
```
|
|
391
|
+
|
|
352
392
|
**POST /crop** is synchronous document QC (contour detection plus one
|
|
353
393
|
SCRFD pass runs in tens of ms): the perspective-corrected canonical
|
|
354
394
|
crop plus quality checks (`document_found`, `aspect_ratio`,
|
|
@@ -596,6 +636,37 @@ record = MRZReader().read("id_card.jpg")
|
|
|
596
636
|
print(record.mrz_type, record.document_number, record.surname)
|
|
597
637
|
```
|
|
598
638
|
|
|
639
|
+
### `BarcodeReader`
|
|
640
|
+
|
|
641
|
+
Defined in [`reading/barcode.py`](src/blitzid/reading/barcode.py);
|
|
642
|
+
requires the `barcode` extra (`pip install blitzid[barcode]`). Decodes
|
|
643
|
+
the PDF417 code on the back of US and Canadian driver's licenses with
|
|
644
|
+
zxing-cpp and parses the AAMVA DL/ID payload (versions 02–14). The
|
|
645
|
+
header's subfile offsets, lengths, and number-of-entries count are
|
|
646
|
+
validated, the mandatory fields are gated, and a failing payload raises
|
|
647
|
+
`BarcodeError` naming the offending field — nothing is invented or
|
|
648
|
+
repaired.
|
|
649
|
+
|
|
650
|
+
| Method | Description |
|
|
651
|
+
|---|---|
|
|
652
|
+
| `read(image_input)` | Decodes and returns a `BarcodeRecord` |
|
|
653
|
+
|
|
654
|
+
**`BarcodeRecord`** — frozen dataclass: `jurisdiction` (the issuer's
|
|
655
|
+
identification number), `aamva_version` (AAMVA version, `"02"`–`"14"`),
|
|
656
|
+
`jurisdiction_version`, `family_name`, `given_name`, `middle_name`,
|
|
657
|
+
`birth_date` / `issue_date` / `expiry_date` (`datetime.date`), `sex`
|
|
658
|
+
(`"M"`/`"F"`/`"X"`), `license_number`, `height` (with its unit, or
|
|
659
|
+
None), `extra_tags` (every un-mapped element kept raw, jurisdiction
|
|
660
|
+
`Z..` extras included), and `symbology` (the decoded barcode symbology,
|
|
661
|
+
stamped by the reader).
|
|
662
|
+
|
|
663
|
+
```python
|
|
664
|
+
from blitzid import BarcodeReader
|
|
665
|
+
|
|
666
|
+
record = BarcodeReader().read("license_back.jpg")
|
|
667
|
+
print(record.aamva_version, record.family_name, record.expiry_date)
|
|
668
|
+
```
|
|
669
|
+
|
|
599
670
|
### `StructuredOCRReader`
|
|
600
671
|
|
|
601
672
|
Defined in [`reading/structurize.py`](src/blitzid/reading/structurize.py);
|
|
@@ -661,6 +732,7 @@ Defined in [`exceptions.py`](src/blitzid/exceptions.py).
|
|
|
661
732
|
| `ModelError` | Model download or load failure |
|
|
662
733
|
| `ImageError` | Image loading, validation, or processing failure |
|
|
663
734
|
| `MRZError` | MRZ not found, malformed, or failed check-digit validation |
|
|
735
|
+
| `BarcodeError` | PDF417 barcode not found, malformed, or failed AAMVA validation |
|
|
664
736
|
| `FaceVerificationError` | Face verification failure (no detectable face, missing landmarks) |
|
|
665
737
|
|
|
666
738
|
Backward-compatibility aliases (`FaceDetectorError`, `ModelDownloadError`, `ImageLoadError`, etc.) are re-exported from `__init__.py`.
|
|
@@ -6,8 +6,9 @@
|
|
|
6
6
|
[](LICENSE)
|
|
7
7
|
|
|
8
8
|
Read identity documents from images: find and compare faces, read printed
|
|
9
|
-
text,
|
|
10
|
-
|
|
9
|
+
text, parse the machine-readable zone, and decode the PDF417 barcode on
|
|
10
|
+
US and Canadian driver's licenses, in one typed Python library that
|
|
11
|
+
runs on an ordinary CPU.
|
|
11
12
|
|
|
12
13
|
No GPU and no CUDA stack: inference runs on onnxruntime's CPU provider,
|
|
13
14
|
face detection takes about 4 ms per image on a laptop, and the weights
|
|
@@ -19,9 +20,10 @@ results in RAM only.
|
|
|
19
20
|
|
|
20
21
|
## Why blitzid
|
|
21
22
|
|
|
22
|
-
- **One install, one API.** Face detection, OCR, MRZ parsing,
|
|
23
|
-
document crop/QC behind a single typed surface,
|
|
24
|
-
a detector, an OCR engine, and your own MRZ code
|
|
23
|
+
- **One install, one API.** Face detection, OCR, MRZ parsing, PDF417
|
|
24
|
+
barcode reading, and document crop/QC behind a single typed surface,
|
|
25
|
+
instead of stitching a detector, an OCR engine, and your own MRZ code
|
|
26
|
+
together.
|
|
25
27
|
- **Validation over guessing.** Every MRZ field is gated by ICAO 9303
|
|
26
28
|
check digits. Obvious OCR confusions (0/O, 1/I, …) are undone by
|
|
27
29
|
each field's alphabet before validation, and a zone that still does
|
|
@@ -44,6 +46,7 @@ results in RAM only.
|
|
|
44
46
|
```bash
|
|
45
47
|
pip install blitzid # or: uv add blitzid
|
|
46
48
|
pip install blitzid[ocr] # with OCR text reading (RapidOCR)
|
|
49
|
+
pip install blitzid[barcode] # with PDF417/AAMVA barcode reading (zxing-cpp)
|
|
47
50
|
pip install blitzid[api] # with the HTTP API server
|
|
48
51
|
```
|
|
49
52
|
|
|
@@ -133,6 +136,15 @@ nationality), and parses TD1 (3x30), TD2 (2x36), and TD3 (2x44) into an
|
|
|
133
136
|
convention (overflow at the start of `optional_data1`, no printed check
|
|
134
137
|
digit).
|
|
135
138
|
|
|
139
|
+
**PDF417 barcode reading.** `BarcodeReader` (the `barcode` extra,
|
|
140
|
+
zxing-cpp) decodes the PDF417 code on the back of US and Canadian
|
|
141
|
+
driver's licenses and parses the AAMVA DL/ID payload into a typed
|
|
142
|
+
`BarcodeRecord` — jurisdiction, AAMVA version, names, dates, sex, and
|
|
143
|
+
license number, plus the raw jurisdiction-specific extras. The header's
|
|
144
|
+
subfile offsets, lengths, and number-of-entries count are validated and
|
|
145
|
+
the mandatory fields are gated, so a malformed payload raises
|
|
146
|
+
`BarcodeError` naming the offending field.
|
|
147
|
+
|
|
136
148
|
**Structured extraction.** `StructuredOCRReader` (same extra) feeds the
|
|
137
149
|
OCR lines to an OpenAI-compatible chat endpoint (e.g. a local vLLM
|
|
138
150
|
server) and returns a typed `StructuredOCR` record; fields the model
|
|
@@ -155,8 +167,8 @@ face). Returns the crop plus a `QualityReport` verdict.
|
|
|
155
167
|
|
|
156
168
|
**HTTP API and Docker.** The `api` extra adds a FastAPI service: an
|
|
157
169
|
async job queue (Redis, persistence disabled) for
|
|
158
|
-
face/attributes/antispoof/OCR/MRZ/structured/consistency,
|
|
159
|
-
`/crop` (document QC), `/verify` (1:1 face comparison), and
|
|
170
|
+
face/attributes/antispoof/OCR/MRZ/barcode/structured/consistency,
|
|
171
|
+
synchronous `/crop` (document QC), `/verify` (1:1 face comparison), and
|
|
160
172
|
`/liveness` (challenge-response liveness), and `/health` for
|
|
161
173
|
orchestration.
|
|
162
174
|
`docker compose up` runs it with all weights baked in.
|
|
@@ -183,9 +195,10 @@ curl http://localhost:8000/api/health
|
|
|
183
195
|
file (any OpenCV-decodable format; PDFs get a precise `415`) and
|
|
184
196
|
`types` (repeated and/or comma-separated: `face`, `attributes`,
|
|
185
197
|
`antispoof`, `ocr`,
|
|
186
|
-
`mrz`, `structured`, `consistency` — `attributes` predicts
|
|
187
|
-
gender, and race per face, `antispoof` scores live vs
|
|
188
|
-
screen photos, `
|
|
198
|
+
`mrz`, `barcode`, `structured`, `consistency` — `attributes` predicts
|
|
199
|
+
age group, gender, and race per face, `antispoof` scores live vs
|
|
200
|
+
printed vs screen photos, `barcode` decodes the PDF417/AAMVA payload
|
|
201
|
+
from a driver's-license back, `structured` rebuilds the OCR text into a
|
|
189
202
|
typed record via the LLM, and `consistency` cross-checks that printed
|
|
190
203
|
record against the machine-readable zone and the portrait's age band
|
|
191
204
|
and gender against the birth date and sex; the photo rows need the
|
|
@@ -193,8 +206,8 @@ and gender against the birth date and sex; the photo rows need the
|
|
|
193
206
|
`BLITZID_LLM_*` config).
|
|
194
207
|
Validation is eager, before the job exists: `422` unknown type or
|
|
195
208
|
missing fields, `400` undecodable bytes or an analysis whose engine is
|
|
196
|
-
unavailable (`ocr`/`mrz` without the `ocr` extra, `
|
|
197
|
-
`consistency` without
|
|
209
|
+
unavailable (`ocr`/`mrz` without the `ocr` extra, `barcode` without the
|
|
210
|
+
`barcode` extra, `structured`/`consistency` without
|
|
198
211
|
the extra or the LLM config), `413` above the
|
|
199
212
|
upload cap, `503` + `Retry-After` when the queue is full or Redis is
|
|
200
213
|
down. Accepted jobs return `202`:
|
|
@@ -207,7 +220,7 @@ curl -F image=@id.jpg -F types=face,ocr localhost:8000/api/analyze
|
|
|
207
220
|
**GET /jobs/{job_id}** polls: `{"status": "queued"}` / `running`; done
|
|
208
221
|
returns `200` with one section per requested type (`face`,
|
|
209
222
|
`attributes`, `antispoof`, `ocr`,
|
|
210
|
-
`mrz`, `structured`, `consistency`; a failing section carries
|
|
223
|
+
`mrz`, `barcode`, `structured`, `consistency`; a failing section carries
|
|
211
224
|
`{"error": ...}` while
|
|
212
225
|
the others still return). The read claims the result, so every later
|
|
213
226
|
read gets `410 Gone`; unknown ids get `404` (an expired job reads `410` until
|
|
@@ -283,6 +296,30 @@ What a done result looks like on the specimen ID (abbreviated):
|
|
|
283
296
|
}
|
|
284
297
|
```
|
|
285
298
|
|
|
299
|
+
On a US driver's-license back, a `barcode` section looks like this:
|
|
300
|
+
|
|
301
|
+
```json
|
|
302
|
+
{
|
|
303
|
+
"record": {
|
|
304
|
+
"jurisdiction": "636014",
|
|
305
|
+
"aamva_version": "08",
|
|
306
|
+
"jurisdiction_version": "00",
|
|
307
|
+
"family_name": "PUBLIC",
|
|
308
|
+
"given_name": "JOHN",
|
|
309
|
+
"middle_name": "QUINCY",
|
|
310
|
+
"birth_date": "1989-01-24",
|
|
311
|
+
"issue_date": "2013-06-04",
|
|
312
|
+
"expiry_date": "2035-01-31",
|
|
313
|
+
"sex": "M",
|
|
314
|
+
"license_number": "D12345678",
|
|
315
|
+
"height": "069 in",
|
|
316
|
+
"extra_tags": {"ZCB": "…"},
|
|
317
|
+
"symbology": "PDF417"
|
|
318
|
+
},
|
|
319
|
+
"processing_time_ms": 87.2
|
|
320
|
+
}
|
|
321
|
+
```
|
|
322
|
+
|
|
286
323
|
**POST /crop** is synchronous document QC (contour detection plus one
|
|
287
324
|
SCRFD pass runs in tens of ms): the perspective-corrected canonical
|
|
288
325
|
crop plus quality checks (`document_found`, `aspect_ratio`,
|
|
@@ -530,6 +567,37 @@ record = MRZReader().read("id_card.jpg")
|
|
|
530
567
|
print(record.mrz_type, record.document_number, record.surname)
|
|
531
568
|
```
|
|
532
569
|
|
|
570
|
+
### `BarcodeReader`
|
|
571
|
+
|
|
572
|
+
Defined in [`reading/barcode.py`](src/blitzid/reading/barcode.py);
|
|
573
|
+
requires the `barcode` extra (`pip install blitzid[barcode]`). Decodes
|
|
574
|
+
the PDF417 code on the back of US and Canadian driver's licenses with
|
|
575
|
+
zxing-cpp and parses the AAMVA DL/ID payload (versions 02–14). The
|
|
576
|
+
header's subfile offsets, lengths, and number-of-entries count are
|
|
577
|
+
validated, the mandatory fields are gated, and a failing payload raises
|
|
578
|
+
`BarcodeError` naming the offending field — nothing is invented or
|
|
579
|
+
repaired.
|
|
580
|
+
|
|
581
|
+
| Method | Description |
|
|
582
|
+
|---|---|
|
|
583
|
+
| `read(image_input)` | Decodes and returns a `BarcodeRecord` |
|
|
584
|
+
|
|
585
|
+
**`BarcodeRecord`** — frozen dataclass: `jurisdiction` (the issuer's
|
|
586
|
+
identification number), `aamva_version` (AAMVA version, `"02"`–`"14"`),
|
|
587
|
+
`jurisdiction_version`, `family_name`, `given_name`, `middle_name`,
|
|
588
|
+
`birth_date` / `issue_date` / `expiry_date` (`datetime.date`), `sex`
|
|
589
|
+
(`"M"`/`"F"`/`"X"`), `license_number`, `height` (with its unit, or
|
|
590
|
+
None), `extra_tags` (every un-mapped element kept raw, jurisdiction
|
|
591
|
+
`Z..` extras included), and `symbology` (the decoded barcode symbology,
|
|
592
|
+
stamped by the reader).
|
|
593
|
+
|
|
594
|
+
```python
|
|
595
|
+
from blitzid import BarcodeReader
|
|
596
|
+
|
|
597
|
+
record = BarcodeReader().read("license_back.jpg")
|
|
598
|
+
print(record.aamva_version, record.family_name, record.expiry_date)
|
|
599
|
+
```
|
|
600
|
+
|
|
533
601
|
### `StructuredOCRReader`
|
|
534
602
|
|
|
535
603
|
Defined in [`reading/structurize.py`](src/blitzid/reading/structurize.py);
|
|
@@ -595,6 +663,7 @@ Defined in [`exceptions.py`](src/blitzid/exceptions.py).
|
|
|
595
663
|
| `ModelError` | Model download or load failure |
|
|
596
664
|
| `ImageError` | Image loading, validation, or processing failure |
|
|
597
665
|
| `MRZError` | MRZ not found, malformed, or failed check-digit validation |
|
|
666
|
+
| `BarcodeError` | PDF417 barcode not found, malformed, or failed AAMVA validation |
|
|
598
667
|
| `FaceVerificationError` | Face verification failure (no detectable face, missing landmarks) |
|
|
599
668
|
|
|
600
669
|
Backward-compatibility aliases (`FaceDetectorError`, `ModelDownloadError`, `ImageLoadError`, etc.) are re-exported from `__init__.py`.
|
|
@@ -14,7 +14,8 @@ layers. Face detection (`src/blitzid/face/`): SCRFD-2.5G ONNX model
|
|
|
14
14
|
an onnxruntime CPU session, with LRU result cache, multi-scale
|
|
15
15
|
detection, NMS + size filtering, batch processing, and bbox
|
|
16
16
|
visualization. Document reading (`src/blitzid/reading/`): RapidOCR
|
|
17
|
-
text lines (`ocr` extra)
|
|
17
|
+
text lines (`ocr` extra), ICAO 9303 MRZ parsing (TD1/TD2/TD3), and
|
|
18
|
+
AAMVA PDF417 driver's-license parsing (`barcode` extra).
|
|
18
19
|
Inputs are file paths, NumPy arrays, or PIL Images; outputs are
|
|
19
20
|
`(x, y, w, h, confidence)` tuples and crops.
|
|
20
21
|
|
|
@@ -22,9 +23,10 @@ Inputs are file paths, NumPy arrays, or PIL Images; outputs are
|
|
|
22
23
|
- python 3.12+ (CI matrix: 3.12 / 3.13 / 3.14), uv, pyproject.toml
|
|
23
24
|
(hatchling backend, src layout, `py.typed`)
|
|
24
25
|
- core deps: numpy, onnxruntime (CPU), opencv-python-headless>=5,
|
|
25
|
-
platformdirs
|
|
26
|
-
- optional extras: `
|
|
27
|
-
|
|
26
|
+
platformdirs, pydantic
|
|
27
|
+
- optional extras: `ocr` (rapidocr, pydantic-ai, langfuse), `barcode`
|
|
28
|
+
(zxing-cpp), `api` (fastapi, uvicorn, redis), `scripts` (tqdm),
|
|
29
|
+
`dev` (pytest, ruff, mypy, pre-commit)
|
|
28
30
|
- Type checker: **mypy** (`strict = true` in pyproject.toml)
|
|
29
31
|
- Lint/format: **ruff** (line-length 88, target py312; E, W, F, I, UP,
|
|
30
32
|
B, SIM, RUF)
|
|
@@ -125,7 +127,8 @@ Write the simplest correct solution. Delete anything that isn't needed.
|
|
|
125
127
|
`src/blitzid/face/` (`detector.py` — `FaceDetectorDNN` plus NMS /
|
|
126
128
|
IoU / size filter; `_face.py` — `Face`, `DetectionMetrics`, cache,
|
|
127
129
|
drawing; `_scrfd.py` — SCRFD preprocess/decode). Document reading in
|
|
128
|
-
`src/blitzid/reading/` (`ocr.py`, `mrz.py`
|
|
130
|
+
`src/blitzid/reading/` (`ocr.py`, `mrz.py`, `barcode.py`,
|
|
131
|
+
`structurize.py`, `consistency.py`, `document.py`). Shared helpers at the
|
|
129
132
|
top level: `_image.py` (input loading and normalization), `_models.py`
|
|
130
133
|
(`ModelManager` — SCRFD ONNX download + CPU
|
|
131
134
|
`onnxruntime.InferenceSession`, default cache dir via `platformdirs`).
|
|
@@ -134,10 +137,12 @@ Write the simplest correct solution. Delete anything that isn't needed.
|
|
|
134
137
|
- Exceptions in `src/blitzid/exceptions.py`: `BlitzIDError` base,
|
|
135
138
|
`ModelError` (model download/load), `ImageError` (load/validate/
|
|
136
139
|
process), `MRZError` (MRZ not found / malformed / failed check-digit
|
|
137
|
-
validation), `
|
|
138
|
-
|
|
139
|
-
|
|
140
|
-
|
|
140
|
+
validation), `BarcodeError` (PDF417 not found / malformed / failed
|
|
141
|
+
AAMVA validation), `FaceVerificationError` (no face for
|
|
142
|
+
verification).
|
|
143
|
+
- Optional deps (PIL, rapidocr via the `ocr` extra, zxing-cpp via the
|
|
144
|
+
`barcode` extra) are lazy-imported inside functions, guarded, with
|
|
145
|
+
targeted `# type: ignore` codes; core must work without them.
|
|
141
146
|
- The MRZ layer (`src/blitzid/reading/mrz.py`) is pure ICAO 9303 logic
|
|
142
147
|
over the OCR reader's text lines — no new models, no guessy repair:
|
|
143
148
|
obvious OCR confusions (0/O, 1/I, 2/Z, 5/S, 6/G, 8/B) resolve
|
|
@@ -161,7 +166,12 @@ Write the simplest correct solution. Delete anything that isn't needed.
|
|
|
161
166
|
surface is `RapidOCRReader.read()` returning `OCRText` records — one
|
|
162
167
|
engine (RapidOCR on onnxruntime CPU), lazy-imported from the `ocr`
|
|
163
168
|
extra. The MRZ surface is `MRZReader.read()` returning an
|
|
164
|
-
`MRZRecord` (same `ocr` extra).
|
|
169
|
+
`MRZRecord` (same `ocr` extra). The barcode surface is
|
|
170
|
+
`BarcodeReader.read()` returning a `BarcodeRecord` (`barcode`
|
|
171
|
+
extra, zxing-cpp decoder; the parser validates the AAMVA header's
|
|
172
|
+
subfile offsets/lengths/count, gates mandatory fields per
|
|
173
|
+
edition-pinned tag maps, keeps unknown tags raw in `extra_tags`,
|
|
174
|
+
and raises `BarcodeError` naming the offending field).
|
|
165
175
|
- `scripts/` are demos, diagnostics, and dataset benchmarks — not part
|
|
166
176
|
of the shipped package; they may use the `scripts` extra.
|
|
167
177
|
- Tests go in `tests/`; validation tests must run without model
|
|
@@ -12,6 +12,7 @@ against a local stack.
|
|
|
12
12
|
|
|
13
13
|
The service quality-checks document photos, reads the printed text and
|
|
14
14
|
the machine-readable code strip, validates the strip's check digits,
|
|
15
|
+
decodes the PDF417 barcode on US and Canadian driver's-license backs,
|
|
15
16
|
compares a selfie to a document portrait, estimates age group, gender,
|
|
16
17
|
and race per face, and cross-checks the copies inside one document.
|
|
17
18
|
|
|
@@ -60,17 +61,18 @@ would trust with that.
|
|
|
60
61
|
|
|
61
62
|
### Check it is healthy
|
|
62
63
|
|
|
63
|
-
`GET /health` returns `200` with
|
|
64
|
-
`attributes`, `antispoof`, `ocr`, `mrz`, `structured`. A
|
|
65
|
-
means that analysis returns `400` with the missing piece
|
|
66
|
-
means the job store is down and the container cannot
|
|
64
|
+
`GET /health` returns `200` with eight model flags: `face`, `verify`,
|
|
65
|
+
`attributes`, `antispoof`, `ocr`, `mrz`, `barcode`, `structured`. A
|
|
66
|
+
`false` flag means that analysis returns `400` with the missing piece
|
|
67
|
+
named. `503` means the job store is down and the container cannot
|
|
68
|
+
accept work.
|
|
67
69
|
|
|
68
70
|
## The blocks
|
|
69
71
|
|
|
70
72
|
| Block | Call | Gives you |
|
|
71
73
|
|---|---|---|
|
|
72
74
|
| Photo quality gate | `POST /crop` | `pass`/`warn`/`reject` verdict, per-check detail, and a corrected crop. Works on documents and selfies alike |
|
|
73
|
-
| Data extraction | `POST /analyze -F types=ocr,mrz,structured` | printed fields, code-strip fields (check digits validated), typed records |
|
|
75
|
+
| Data extraction | `POST /analyze -F types=ocr,mrz,barcode,structured` | printed fields, code-strip fields (check digits validated), the PDF417/AAMVA record from a license back, typed records |
|
|
74
76
|
| Age signal | `POST /analyze -F types=attributes` | age band, gender, and race per face |
|
|
75
77
|
| Spoof signal | `POST /analyze -F types=antispoof` | per-face scores: live, printed photo, screen photo |
|
|
76
78
|
| Document self-check | `POST /analyze -F types=consistency` | printed text vs code strip vs portrait, per-field verdicts |
|
|
@@ -119,8 +121,9 @@ One document, both sides where the card has a back:
|
|
|
119
121
|
- **Passport**: one photo, the data page. The code strip is at the
|
|
120
122
|
bottom of the same page.
|
|
121
123
|
- **ID card**: front and back. The back carries the code strip.
|
|
122
|
-
- **Driver license**: front and back. The back carries the
|
|
123
|
-
|
|
124
|
+
- **Driver license**: front and back. The back carries the PDF417
|
|
125
|
+
barcode with the AAMVA record (a machine copy of the printed data,
|
|
126
|
+
validated by the `barcode` analysis).
|
|
124
127
|
|
|
125
128
|
Capture paths depend on what the user must prove. "Selfie, then ID
|
|
126
129
|
front and back" is the base. "Selfie, then passport" is the same flow
|
|
@@ -163,7 +166,8 @@ Pick types per document:
|
|
|
163
166
|
| Passport data page | `face,ocr,mrz,structured,consistency` |
|
|
164
167
|
| ID card (back photographed) | `face,ocr,mrz,structured,consistency` |
|
|
165
168
|
| ID card (front only) | `face,ocr,structured` (no code strip to read) |
|
|
166
|
-
| Driver license | `face,ocr,structured,
|
|
169
|
+
| Driver license (back photographed) | `face,ocr,mrz,barcode,structured,consistency` |
|
|
170
|
+
| Driver license (front only) | `face,ocr,structured,attributes` (no barcode or MRZ on the front) |
|
|
167
171
|
| Vehicle registration / insurance | `ocr,structured` |
|
|
168
172
|
| Plate photo | `structured` |
|
|
169
173
|
|
|
@@ -272,10 +276,11 @@ the two crops, not a boolean.
|
|
|
272
276
|
|
|
273
277
|
Read this before you promise a flow:
|
|
274
278
|
|
|
275
|
-
- **Barcode
|
|
276
|
-
|
|
277
|
-
|
|
278
|
-
|
|
279
|
+
- **Barcode-vs-print cross-check.** The `barcode` analysis decodes the
|
|
280
|
+
PDF417/AAMVA record on US and Canadian license backs, but the
|
|
281
|
+
`consistency` section does not compare it against the printed front
|
|
282
|
+
yet — the strongest tamper signal those cards have. It compares the
|
|
283
|
+
code strip (when the card has one) today.
|
|
279
284
|
- **Video replay and generated faces.** The liveness block stops
|
|
280
285
|
static spoofs and the `antispoof` block flags cameras pointed at
|
|
281
286
|
photos and displays. Filming a screen that plays a live video, or a
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
[project]
|
|
2
2
|
name = "blitzid"
|
|
3
|
-
version = "0.3.
|
|
3
|
+
version = "0.3.2"
|
|
4
4
|
description = "Modular DNN-based ID document reading framework (face detection, OCR, MRZ) optimized for CPU deployment"
|
|
5
5
|
readme = "README.md"
|
|
6
6
|
requires-python = ">=3.12"
|
|
@@ -47,6 +47,11 @@ ocr = [
|
|
|
47
47
|
# Optional: Langfuse tracing of the LLM calls (env-gated).
|
|
48
48
|
"langfuse",
|
|
49
49
|
]
|
|
50
|
+
barcode = [
|
|
51
|
+
# PDF417 decoding for AAMVA driver's-license payloads (also the
|
|
52
|
+
# encoder the test suite round-trips synthetic payloads with).
|
|
53
|
+
"zxing-cpp",
|
|
54
|
+
]
|
|
50
55
|
api = [
|
|
51
56
|
"fastapi",
|
|
52
57
|
"uvicorn[standard]",
|
|
@@ -74,7 +79,7 @@ dev = [
|
|
|
74
79
|
"fakeredis",
|
|
75
80
|
]
|
|
76
81
|
all = [
|
|
77
|
-
"blitzid[api,ocr,scripts,dev]",
|
|
82
|
+
"blitzid[api,barcode,ocr,scripts,dev]",
|
|
78
83
|
]
|
|
79
84
|
|
|
80
85
|
# ── Build system ───────────────────────────────────────────
|
|
@@ -134,6 +139,7 @@ module = [
|
|
|
134
139
|
"PIL.*",
|
|
135
140
|
"rapidocr",
|
|
136
141
|
"rapidocr.*",
|
|
142
|
+
"zxingcpp",
|
|
137
143
|
"pydantic_ai",
|
|
138
144
|
"pydantic_ai.*",
|
|
139
145
|
"langfuse",
|
|
@@ -15,6 +15,9 @@ Public API
|
|
|
15
15
|
- ``OCRText`` — recognized text line (bbox, text, confidence).
|
|
16
16
|
- ``MRZReader`` — the ICAO 9303 machine-readable zone reader (``ocr`` extra).
|
|
17
17
|
- ``MRZRecord`` — parsed MRZ fields.
|
|
18
|
+
- ``BarcodeReader`` — the AAMVA PDF417 driver's-license reader
|
|
19
|
+
(``barcode`` extra).
|
|
20
|
+
- ``BarcodeRecord`` — parsed AAMVA payload fields.
|
|
18
21
|
- ``StructuredOCRReader`` — LLM-backed structured output from OCR lines
|
|
19
22
|
(``ocr`` extra).
|
|
20
23
|
- ``StructuredOCR``, ``ExtractedField`` — structured-output schemas.
|
|
@@ -25,10 +28,11 @@ Public API
|
|
|
25
28
|
- ``DocumentCropper`` — document localization, perspective crop, and QC.
|
|
26
29
|
- ``QualityReport`` — document crop quality-check verdict.
|
|
27
30
|
- ``BlitzIDError``, ``ModelError``, ``ImageError``, ``MRZError``,
|
|
28
|
-
``FaceVerificationError`` — exception hierarchy.
|
|
31
|
+
``BarcodeError``, ``FaceVerificationError`` — exception hierarchy.
|
|
29
32
|
"""
|
|
30
33
|
|
|
31
34
|
from blitzid.exceptions import (
|
|
35
|
+
BarcodeError,
|
|
32
36
|
BlitzIDError,
|
|
33
37
|
FaceVerificationError,
|
|
34
38
|
ImageError,
|
|
@@ -40,6 +44,7 @@ from blitzid.face._attributes import FaceAttributeReader, FaceAttributes
|
|
|
40
44
|
from blitzid.face._face import DetectionMetrics, Face, VerificationResult
|
|
41
45
|
from blitzid.face.detector import FaceDetectorDNN
|
|
42
46
|
from blitzid.face.verifier import FaceVerifier
|
|
47
|
+
from blitzid.reading.barcode import BarcodeReader, BarcodeRecord
|
|
43
48
|
from blitzid.reading.consistency import (
|
|
44
49
|
ConsistencyReport,
|
|
45
50
|
FieldComparison,
|
|
@@ -64,6 +69,9 @@ InvalidParameterError = BlitzIDError
|
|
|
64
69
|
__all__ = [
|
|
65
70
|
"AntiSpoofReader",
|
|
66
71
|
"AntiSpoofResult",
|
|
72
|
+
"BarcodeError",
|
|
73
|
+
"BarcodeReader",
|
|
74
|
+
"BarcodeRecord",
|
|
67
75
|
"BlitzIDError",
|
|
68
76
|
"ConsistencyReport",
|
|
69
77
|
"DetectionMetrics",
|
|
@@ -1,4 +1,10 @@
|
|
|
1
|
-
"""Date coercion shared by the reading pipeline.
|
|
1
|
+
"""Date coercion shared by the reading pipeline.
|
|
2
|
+
|
|
3
|
+
Alongside the pipeline's ISO 8601 and ``DD.MM.YYYY`` shapes, the AAMVA
|
|
4
|
+
barcode reader feeds MMDDYYYY (U.S.) and CCYYMMDD (Canada) date
|
|
5
|
+
elements through these formats; ``%m%d%Y`` is tried first so a string
|
|
6
|
+
valid in both shapes (a US date) keeps the US reading.
|
|
7
|
+
"""
|
|
2
8
|
|
|
3
9
|
from __future__ import annotations
|
|
4
10
|
|
|
@@ -7,7 +13,7 @@ from datetime import datetime
|
|
|
7
13
|
|
|
8
14
|
__all__ = ["parse_date"]
|
|
9
15
|
|
|
10
|
-
_DATE_FORMATS = ("%Y-%m-%d", "%d.%m.%Y")
|
|
16
|
+
_DATE_FORMATS = ("%Y-%m-%d", "%d.%m.%Y", "%m%d%Y", "%Y%m%d")
|
|
11
17
|
|
|
12
18
|
|
|
13
19
|
def parse_date(value: object) -> object:
|