Sythos Barcode Suite

Read and write barcodes in JavaScript.

Original Sythos implementation, zero runtime dependencies, MIT. It runs unmodified in Node, in browsers (including Safari on iOS) and in web workers. The core requires no canvas, no filesystem and no DOM — images go in and come out as plain { data, width, height } RGBA objects, which is exactly what an ImageData is.

The code is complete and entirely human-readable. The full source ships. There is no WebAssembly, no native addon, no compiled artefact, no binary blob and no minified file anywhere in this repository — every tracked file is text you can open and read.

That includes the prebuilt bundles. They are generated, concatenated and wrapped from src/ by the project's own bundler, but nothing is stripped in the process: bundle/sythos-barcode.js runs to roughly 7,900 lines, about a third of them comments, averaging a little over 30 characters a line. Open it anywhere and you are reading the same annotated code as the source, in the same order — a convenience, not a black box.

Don't take that on trust either; it takes one command:

awk '{ n += length($0) } END { print "lines:", NR, " avg length:", int(n/NR) }' bundle/sythos-barcode.js

A minified bundle gives you a handful of lines averaging thousands of characters. This one does not, and that is the whole point.

This is deliberate. A barcode library decides what a scanner believes a label says, so it belongs in the category of code you can audit rather than have to trust. Every constant table, every check digit and every error-correction step is here in full, with the reasoning next to it.

import { encode, decode, toSVG, toImageData } from './src/index.js';

const matrix = encode('https://example.com', { format: 'qr', ecc: 'M' });
const svg    = toSVG(matrix, { scale: 8 });

const found = decode(toImageData(matrix, { scale: 4 }), { formats: ['qr'] });
console.log(found[0].text);   // 'https://example.com'

---

Quick start

There are four ways in, and none of them needs a build step.

1. npm

npm install @sythos/js_barcode_universal

yarn add @sythos/js_barcode_universal and pnpm add @sythos/js_barcode_universal do the same thing. Nothing is installed alongside it — there are no runtime dependencies, no postinstall script and no native build. The package is plain ESM ("type": "module") and asks for Node 18 or newer.

import { encode, decode, toSVG, toImageData } from '@sythos/js_barcode_universal';

const code = encode('SYT-2026-0042', { format: 'code128' });

const svg = toSVG(code, { scale: 2, margin: 10, barHeight: 60 });
// '<svg xmlns="http://www.w3.org/2000/svg" width="374" height="100" …'

const found = decode(toImageData(code, { scale: 4, margin: 10 }), { formats: ['code128'] });
console.log(found[0].text);   // 'SYT-2026-0042'

Subpath exports hand you one layer instead of the whole surface, which is what lets a tree-shaking bundler drop everything you did not ask for. Importing only the QR writer and only the SVG renderer never pulls in the 1D formats, the PNG encoder or the read pipeline:

import { encodeQR } from '@sythos/js_barcode_universal/qr';
import { toSVG }    from '@sythos/js_barcode_universal/render/svg';

const svg = toSVG(encodeQR('https://example.com', { ecc: 'M' }), { scale: 8 });
// a 25×25 module symbol — 264×264 px at scale 8 with the default 4-module quiet zone

Subpath	What it exports
@sythos/js_barcode_universal	The whole surface: encode, decode, every renderer, every error type
@sythos/js_barcode_universal/core	BitMatrix, GaloisField, Reed–Solomon, the error classes
@sythos/js_barcode_universal/image	LuminanceSource, the binarizers, grid sampling, PerspectiveTransform
@sythos/js_barcode_universal/oned	The per-format 1D writers (encodeEAN13, encodeCode128, …) and decodeOneD
@sythos/js_barcode_universal/qr	encodeQR, decodeQR, detectQR, detectAndDecodeQR
@sythos/js_barcode_universal/datamatrix	encodeDataMatrix, decodeDataMatrix, detectDataMatrix, detectAndDecodeDataMatrix
@sythos/js_barcode_universal/aztec	encodeAztec, decodeAztec, detectAztec, detectAndDecodeAztec
@sythos/js_barcode_universal/pdf417	encodePDF417, decodePDF417, detectPDF417, detectAndDecodePDF417
@sythos/js_barcode_universal/micropdf417	encodeMicroPDF417, decodeMicroPDF417, detectMicroPDF417, detectAndDecodeMicroPDF417
@sythos/js_barcode_universal/render	Every renderer plus isWebGL2Available / isWebGPUAvailable
@sythos/js_barcode_universal/render/svg	toSVG, toSVGDataURI
@sythos/js_barcode_universal/render/png	toPNG, toPNGDataURI
@sythos/js_barcode_universal/render/image-data	toImageData, toCanvas
@sythos/js_barcode_universal/bundle	The prebuilt ESM bundle, as one file
@sythos/js_barcode_universal/bundle/iife	The prebuilt IIFE bundle, for a <script> tag

The unpkg and jsdelivr fields point at the IIFE bundle, so a CDN needs no install at all:

<script src="https://unpkg.com/@sythos/js_barcode_universal"></script>
<script src="https://unpkg.com/@sythos/js_barcode_universal@1.2.5"></script>
<script src="https://cdn.jsdelivr.net/npm/@sythos/js_barcode_universal@1.2.5"></script>

Pin the version for anything you ship; the unpinned form resolves to latest and will move under you.

Both CDNs serve the same file the repository ships in bundle/sythos-barcode.js — byte for byte, since that is exactly what npm publishes.

2. A <script> tag

bundle/sythos-barcode.js is a self-contained IIFE that exposes a single global, SythosBarcode. It works straight from file:// — open an HTML file off your disk and it runs.

<script src="bundle/sythos-barcode.js"></script>
<script>
  var encode = SythosBarcode.encode;
  var toSVGDataURI = SythosBarcode.toSVGDataURI;

  var img = new Image();
  img.src = toSVGDataURI(encode('https://example.com', { format: 'qr' }), { scale: 8 });
  document.body.appendChild(img);
</script>

3. ESM bundle

bundle/sythos-barcode.esm.js is the same code as a single ES module, for <script type="module">, a bundler, or Node.

import { encode, toSVG, toPNG } from './bundle/sythos-barcode.esm.js';

const ean = encode('5901234123457', { format: 'ean13' });

const svg = toSVG(ean, { scale: 3, margin: 10, barHeight: 80 });
// '<svg xmlns="http://www.w3.org/2000/svg" width="345" height="141" …'

toPNG(ean, { scale: 3, barHeight: 80 }).then((bytes) => {
  // Uint8Array — a 1-bit palette PNG
});

4. The source directly

src/index.js is plain ESM with JSDoc types and no build step of its own. Import it and let your bundler tree-shake; the package is marked side-effect free.

import { encode, decode, toImageData, listFormats } from './src/index.js';

const matrix = encode('https://example.com', { format: 'qr', ecc: 'M' });
const image  = toImageData(matrix, { scale: 4, margin: 4 });

const found = decode(image, { formats: ['qr'] });
// [ { text: 'https://example.com', format: 'qr', version: 2, ecc: 'M', … } ]

Decoding takes anything ImageData-shaped, so a canvas, an OffscreenCanvas, createImageBitmap, or an image library's raw buffer all satisfy it without an adapter:

const ctx = canvas.getContext('2d');
ctx.drawImage(video, 0, 0, canvas.width, canvas.height);

for (const hit of decode(ctx.getImageData(0, 0, canvas.width, canvas.height))) {
  console.log(hit.format, hit.text);
}

decode returns an array, empty when nothing is found. A frame with no barcode is an ordinary outcome for a camera loop, not an error, so the common case needs no try/catch. Use decodeStrict when absence really is a failure.

---

Supported formats

Generated from listFormats(), which reports writing and reading as separate capabilities. Writing a symbology is a table lookup; reading one needs a detector that finds it in a photograph. The two lists legitimately differ, and saying so here is better than failing at call time.

Format	id	Kind	Write	Read
EAN-13	ean13	1D	✅	✅
EAN-8	ean8	1D	✅	✅
UPC-A	upca	1D	✅	✅
UPC-E	upce	1D	✅	✅
ISBN (Bookland)	isbn	1D	✅	✅ 1
Code 128	code128	1D	✅	✅
GS1-128	gs1128	1D	✅	✅ 1
Code 39	code39	1D	✅	✅
Code 93	code93	1D	✅	✅
ITF (Interleaved 2 of 5)	itf	1D	✅	✅
ITF-14	itf14	1D	✅	✅ 1
Codabar	codabar	1D	✅	✅
Code 11	code11	1D	✅	—
MSI Plessey	msi	1D	✅	—
Pharmacode	pharmacode	1D	✅	—
QR Code	qr	2D	✅	✅
Data Matrix ECC 200	datamatrix	2D	✅	✅
Aztec Code	aztec	2D	✅	✅
PDF417	pdf417	2D	✅	✅
MicroPDF417	micropdf417	2D	✅	✅

Twenty formats, all writable, seventeen readable. Code 11, MSI Plessey and Pharmacode remain write-only in the generic image pipeline. PDF417 exposes direct matrix decoding, automatic camera localization and an assisted quadrilateral sampler through its subpath. Its detector is validated on degraded synthetic photographs and real Pixel 10/Chrome and iPhone 17/Safari camera tests; external black-box vectors from ZXing 3.5.3 and bwip-js also pass in both directions. Text and Numeric vectors are covered bidirectionally; binary byte-for-byte interop remains explicitly unclaimed until a dedicated external byte corpus is added.

Data Matrix ECC 200

datamatrix writes and reads the 30 classic ECC 200 square and rectangular symbol sizes. The encoder supports ASCII compaction (including numeric pairs), Base256 binary payloads, automatic or forced square/rectangular shape, Reed–Solomon error correction and GS1 FNC1 in the first position. DMRE is not included.

import { encodeDataMatrix, decodeDataMatrix } from '@sythos/js_barcode_universal/datamatrix';

const symbol = encodeDataMatrix('0101234567890128', { gs1: true, shape: 'square' });
const result = decodeDataMatrix(symbol);
console.log(result.text, result.gs1); // 0101234567890128 true

Binary content is accepted as a Uint8Array with encoding: 'base256'. The current high-level decoder handles ASCII and Base256 codewords; C40, Text, X12 and EDIFACT input symbols are not yet decoded. The current detector accepts axis-aligned square or rectangular symbols; the normal decode(image, { formats: ['datamatrix'] }) pipeline also retries quarter-turn rotations. Arbitrary-angle and perspective-skewed Data Matrix photographs are not yet guaranteed.

Aztec Code

aztec writes Compact layers 1–4 and Full layers 1–32, selecting a fitting symbol automatically unless layers and compact are forced. It supports the five Aztec text tables and UTF-8 byte payloads through Binary Shift, with eccPercent (default 23) controlling the requested error-correction level. ECI is not yet a configurable public option.

import { encodeAztec, decodeAztec } from '@sythos/js_barcode_universal/aztec';

const symbol = encodeAztec('Greetings My Lord Sythos  👋', { eccPercent: 23 });
const result = decodeAztec(symbol);
console.log(result.text); // Greetings My Lord Sythos  👋

The image detector handles rotation, inverted polarity and quadrilateral sampling around the central bull’s-eye. Severe photographic perspective remains an interoperability and robustness gate rather than a guaranteed capability.

PDF417 (writer, matrix decoder and camera reader)

pdf417 supports PDF417 Text, Byte and Numeric compaction, ECI 3 (ISO-8859-1) and ECI 26 (UTF-8), ECC levels 0–8, row-height inference and Reed–Solomon erasure correction. The direct matrix decoder is available from @sythos/js_barcode_universal/pdf417.

The image helper handles clean module-aligned raster symbols, integer scale, right-angle rotations, automatic perspective estimation, mild blur/noise and an application-supplied quadrilateral. Results expose bytes and ordered segments for byte-preserving payloads. Real device validation covers Pixel 10/Chrome and iPhone 17/Safari with printed symbols and continuous camera capture. Extreme glare, severe occlusion, curved media and multi-symbol scenes remain outside the validated robustness envelope.

import { encodePDF417, decodePDF417 } from '@sythos/js_barcode_universal/pdf417';

const symbol = encodePDF417('AAMVA SAMPLE', { eccLevel: 3 });
console.log(decodePDF417(symbol).text);

MicroPDF417

micropdf417 writes and reads the 34 fixed MicroPDF417 variants. It supports Text, Byte and Numeric compaction, plus Byte-compaction ECI 3 (ISO-8859-1) and 26 (UTF-8). columns, rowHeight and aspectRatio let callers constrain automatic variant selection.

import { encodeMicroPDF417, decodeMicroPDF417 } from '@sythos/js_barcode_universal/micropdf417';

const symbol = encodeMicroPDF417('MICRO PDF417', { compaction: 'text' });
console.log(decodeMicroPDF417(symbol).text);

The detector accepts clean, integer-scaled raster symbols and quarter-turn rotations. Arbitrary perspective, severe photographic degradation and multi-symbol scenes are not yet claimed as robust capabilities.

Not implemented

GS1 DataBar and MaxiCode are not implemented — neither writing nor reading. Data Matrix ECC 200 is implemented for its classic square and rectangular symbols; DMRE remains outside the current scope. See PLAN.md for the remaining symbologies.

---

Live examples

Two self-contained pages, each loading the IIFE bundle with a plain <script> tag. Both open directly from disk — double-click the file, no server and no build. This examples/ directory is the single canonical source; the development workspace references these files instead of keeping a second copy.

examples/create.html

Pick any writable format, type a payload, and watch the symbol redraw as you type; download it as PNG or SVG. For QR it adds a content-type builder that assembles the payload for you across URL, email, phone, SMS, Wi-Fi network, contact card (both vCard and MeCard), geo location and calendar event — with correct escaping for each — and shows you the exact string it produced, so you can see what a Wi-Fi or vCard QR actually contains. ECC level, version, scale, margin and both colours are exposed.

examples/read.html

Decode from an image: drop one onto the page, or click to choose a file. It then offers a live camera loop that decodes continuously from the video stream.

The file and drag-drop path works anywhere, file:// included. The camera needs http(s), because getUserMedia requires a secure context and refuses to run from file://. Serve the folder over localhost for that half; the page detects the situation and says so rather than failing silently.

---

API summary

Two functions carry the whole surface. Everything else is a renderer or a format-specific escape hatch.

Encoding and decoding

encode(text, options?) → BitMatrix

options: format (default 'qr'), ecc ('L'|'M'|'Q'|'H'), version (QR 1–40, auto if omitted), checkDigit, fullAscii (Code 39 extended), gs1 (emit a leading FNC1). Data Matrix ECC 200 accepts shape: 'any' | 'square' | 'rectangular' and encoding: 'ascii' | 'base256'. Aztec accepts layers, compact and eccPercent; it transports UTF-8 byte payloads through Binary Shift, and it does not expose configurable ECI yet. MicroPDF417 accepts compaction: 'auto' | 'text' | 'byte' | 'numeric', ECI 3 or 26 for Byte compaction, and optional columns, rowHeight and aspectRatio constraints.

encode('5901234123457', { format: 'ean13' })
encode('ABC-123', { format: 'code39', fullAscii: true, checkDigit: true })
encode('https://example.com', { format: 'qr', ecc: 'H', version: 7 })
encode('0101234567890128', { format: 'datamatrix', gs1: true })
encode('Greetings My Lord Sythos  👋', { format: 'aztec', eccPercent: 23 })
encode('MICRO PDF417', { format: 'micropdf417', compaction: 'text' })

decode(image, options?) → Result[]
decodeStrict(image, options?) → Result        // throws NotFoundError instead of returning []

image is { data, width, height } with RGBA bytes. options: formats (restrict the search, and go faster), tryHarder (retry inverted, default true), binarizer ('global' | 'hybrid' | 'auto'). A Result carries at least text and format; QR results also carry bytes, version and ecc.

listFormats() → { id, label, canWrite, canRead, kind }[]

The table above is this function's output. Read it at runtime rather than hard-coding a format list — that is how the demo pages build their dropdowns.

Renderers

toSVG(matrix, options?) → string                       // one merged <path>, not a rect per module
toSVGDataURI(matrix, options?) → string                // data: URI for an <img src>
toImageData(matrix, options?) → { data, width, height }
toPNG(matrix, options?) → Promise<Uint8Array>          // 1-bit palette PNG
toPNGDataURI(matrix, options?) → Promise<string>
toCanvas(matrix, canvas, options?) → boolean           // 2D context
renderToCanvasAuto(matrix, canvas, options?) → { backend: 'webgl2' | '2d' | 'none' }
renderToCanvasAutoAsync(matrix, canvas, options?) → Promise<{ backend: 'webgpu' | 'webgl2' | '2d' | 'none' }>

The two PNG functions are async because they use the platform's deflate — node:zlib or CompressionStream — and fall back to stored blocks where neither exists.

renderToCanvasAuto is synchronous and therefore cannot reach WebGPU: acquiring an adapter is asynchronous, and a synchronous function can never wait for one. Use renderToCanvasAutoAsync when you want WebGPU in the chain. Both fall through to the 2D context, which always exists.

All renderers share the same options:

Option	Default	Meaning
scale	8	Pixels per module
margin	4	Quiet-zone modules on every side
dark	'#000000'	Colour of set modules
light	'#ffffff'	Colour of clear modules; 'none' for transparent
barHeight	auto	1D only: total bar height in pixels

Also exported: BitMatrix, the error types (BarcodeError, EncodeError, NotFoundError, FormatError, ChecksumError), the per-format writers (encodeEAN13, encodeCode128, …), the QR entry points (encodeQR, decodeQR, detectQR, detectAndDecodeQR), the image primitives (LuminanceSource, binarize, binarizeGlobal, binarizeHybrid), and the capability probes isWebGL2Available / isWebGPUAvailable.

---

How it works

BitMatrix is the interchange type. Every writer produces one, every reader consumes one, every renderer draws one. That single currency is what keeps symbologies and output targets independent of each other — adding a format touches no renderer, and adding a renderer touches no format. Storage is row-packed into a Uint32Array: one allocation, cache-friendly row scans, and cheap whole-row operations for the 1D readers.

A set bit is a dark module. This matches how every specification describes its symbols; renderers invert where their medium needs it.

encode returns no quiet zone. The margin is a rendering decision, not an encoding one — how much white space a symbol needs depends on where it is going — so the renderers add it and the matrix stays the pure symbol. For the same reason, linear symbols come back exactly one module tall: height carries no information in a 1D barcode, so encoding one would be inventing data. The renderer stretches the single row to barHeight before applying the quiet zone, so the margin ends up uniform on all four sides.

The read pipeline is a straight line, each stage a separate module:

RGBA bytes → luminance → binarize → detect → sample → error-correct → decode

Luminance conversion flattens the image to greyscale. Binarization turns that into a BitMatrix, either globally or with a hybrid local threshold that survives uneven lighting. Detection locates a symbol and its corners in that bit plane. Detectors that recover four perspective-aware corners (currently QR) sample the symbol back through a perspective transform. Data Matrix currently uses an axis-aligned bounding box plus quarter-turn retries. Error correction repairs what the camera lost. Only then is the payload decoded.

Reed–Solomon is generic over the finite field. The GaloisField class is constructed with a field order and a primitive polynomial rather than hard-coding GF(256), which is what lets one implementation serve QR, and the prime field GF(929) that PDF417 needs. Prime fields are the subtle case: in a binary field addition and subtraction are both XOR, so a decoder that inlines ^ for field addition passes every binary field and fails only the prime one.

---

About the GPU path

The WebGL2 and WebGPU backends accelerate drawing a barcode, not computing one. That is worth stating plainly, because "GPU barcode generation" naturally suggests the latter.

Encoding is sequential integer work: Reed–Solomon polynomial division, mask penalty scoring, bit placement along a zig-zag path. Each step depends on the one before it, which is precisely the shape a GPU cannot exploit. A complete QR encode takes well under a millisecond on the CPU — less time than dispatching a compute shader and reading the result back would cost. Moving it to the GPU would make it slower.

So encoding stays on the CPU because that is the correct engineering answer, not because something is missing. Where the GPU genuinely earns its place is drawing large symbols, or many symbols per frame, straight into a canvas without a CPU-side pixel buffer — and it would earn it again on the read side, where per-frame greyscale conversion and block statistics over a 4K camera image are both the real bottleneck and embarrassingly parallel.

---

Browser support

The syntax floor is iOS Safari 15. No Array.prototype.at, no top-level await, no Object.groupBy.

OffscreenCanvas, WebGL2 and WebGPU are feature-detected, never assumed. The 2D canvas path always exists, so nothing is unreachable on an older device — renderToCanvasAuto degrades to it and reports which backend actually drew.

---

Licence

MIT © 2026 Sythos. Every source file carries the header.

The implementation is original Sythos work. No third-party barcode source code is copied into or shipped by this package. The symbologies are implemented from published descriptions of the formats and from original Sythos data structures; MicroPDF417 normative/public values carry provenance and pending legal review in NOTICE.md. The distributed package has no runtime third-party licence or co-author attribution.

Trademark is not licence. QR Code® is a registered trademark of DENSO WAVE; Aztec Code, MaxiCode and GS1 DataBar are likewise marks of their owners. A trademark does not restrict implementing a symbology, but it does constrain branding — which is why this package is named descriptively rather than after any mark.

Data Matrix ECC 200 is governed by ISO/IEC 16022:2024; GS1 DataMatrix additionally uses the GS1 General Specifications and a leading FNC1. Its engineering provenance, patent and trademark research notes are recorded in licenses/data-matrix.license, with unresolved claims kept explicitly marked using scoped review labels.

PDF417 and MicroPDF417 provenance and legal review notes are recorded in licenses/pdf417.license, licenses/micropdf417.license and the attribution log in NOTICE.md.

LICENSE carries the full MIT text plus an informational appendix inventorying the specification copyrights, patent history and trademarks that surround these symbologies. None of them is resolved by this file; the appendix is an engineering inventory, not legal advice. NOTICE.md records the origin of the code and how its correctness is verified.

---

Contributing and roadmap

PLAN.md is the live status document: what is shipped, what is next, and the ground rules — chief among them that no third-party source code is copied into or shipped by this project, while public or normative values remain provenance-tracked and subject to legal review.

Issues and pull requests are welcome at Sythos/JS_Barcode_Universal. A patch that adds a symbology should implement it from the published description of the format, generate its tables where they are derivable, and come with a symbol that a scanner this project did not write has actually read — that last one is the check that matters.

Footnotes

1. gs1128, itf14 and isbn are sub-variants that share a decoder with their base format, so they decode under that base id: a GS1-128 comes back as code128, an ITF-14 as itf, and an ISBN as ean13. The payload is intact either way — a GS1-128 is a Code 128 with a leading FNC1, an ITF-14 is an ITF fixed at fourteen digits, and an ISBN barcode is an EAN-13 with a 978/979 prefix. Match on result.format === 'code128' rather than 'gs1128', or a condition on the sub-variant id will silently never fire. ↩ ↩² ↩³