native-iqa

No-reference image quality assessment for Node.js. Runs BRISQUE feature extraction plus blur, exposure, and noise detection in Rust via N-API. All processing happens off the JS thread.

Accepts JPEG, PNG, WebP, BMP, and TIFF. Returns scores in under 10ms for typical web images.

npm install native-iqa

Requires Node.js 18 or later. No Rust toolchain needed.

Usage

const iqa = require('native-iqa');
const fs = require('fs');

const buffer = fs.readFileSync('photo.jpg');
const result = await iqa.analyze(buffer);
// {
//   score: 28.4,
//   quality: 'good',
//   isBlurry: false,
//   exposure: 'normal',
//   noiseLevel: 'low',
//   warnings: []
// }

API

analyze(buffer)

Full quality analysis. Returns a Promise.

const result = await iqa.analyze(buffer);

result.score        number    Quality score 0-100. Lower is better.
result.quality      string    'excellent' | 'good' | 'acceptable' | 'poor'
result.isBlurry     boolean   True when Laplacian variance is below threshold
result.exposure     string    'normal' | 'underexposed' | 'overexposed'
result.noiseLevel   string    'low' | 'medium' | 'high'
result.warnings     string[]  Human-readable list of detected issues

Score bands:

0-20    excellent
20-40   good
40-60   acceptable
60+     poor

The score is computed by a pre-trained SVR model using 36 BRISQUE features extracted from MSCN coefficients across two scales. Natural, sharp photos score low. Flat, blurry, or noisy images score high.

quickScore(buffer)

Returns only the numeric score. Faster than analyze because it skips blur, exposure, and noise detection entirely.

const score = await iqa.quickScore(buffer);
// 28.4

validate(buffer, options)

Returns a pass/fail result against configurable thresholds.

const result = await iqa.validate(buffer, {
  maxScore: 50,
  rejectBlurry: true,
  rejectBadExposure: true,
});
// { passed: true, score: 28.4, quality: 'good', failures: [] }

Options:

maxScore           number    Reject if score exceeds this value. Default 60.
rejectBlurry       boolean   Reject blurry images. Default false.
rejectBadExposure  boolean   Reject underexposed or overexposed images. Default false.

scoreBatch(buffers)

Analyze multiple images in parallel. Uses a Rayon thread pool internally so throughput scales with available cores. Results are returned in the same order as input.

const results = await iqa.scoreBatch([bufferA, bufferB, bufferC]);

Each item in the result array has the same shape as analyze().

Supported platforms

Prebuilt binaries are provided for:

• macOS arm64 (Apple Silicon)
• macOS x64
• Linux x64 (glibc)
• Linux arm64 (glibc)
• Windows x64

Building from source

Requires Rust 1.88 or later and the napi-rs CLI.

npm run build
npm test

Benchmark

npm run bench

Runs analyze, quickScore, and scoreBatch at 256, 512, and 1024px over 50 iterations and prints median and p95 latency. No extra dependencies required.

Accuracy

BRISQUE scores match OpenCV's QualityBRISQUE (LIVE model) within ±1 point across a representative test set covering sharp, blurry, noisy, dark, and bright images. The small residual (~0.5 mean offset) is due to float64 vs float32 precision.

License

MIT