wasmosaic
v0.2.0
Published
Browser image processing primitives written in Rust and exposed through WebAssembly and TypeScript.
Maintainers
seyamalamReadme
WASMosaic
wasmosaic is an experimental browser image-processing package written in Rust and TypeScript. Rust owns the pixel loops. WebAssembly carries them into the browser. The TypeScript layer validates inputs and provides an API that works with ImageData.
The published 0.1.0 release provides four RGBA operations. Current development also includes Rust-owned typed matrices, core matrix operations, and the first OpenCV.js-compatible cvtColor slice. The long-term target is full parity with the pinned OpenCV.js 4.13.0 browser bindings, followed by typed browser adapters that OpenCV.js does not provide.
Install
bun add wasmosaicThe npm equivalent is npm install wasmosaic.
Browser example
import { imageDataFromRgbaImage, initOpenCv, rgbaImageFromImageData } from "wasmosaic";
const canvas = document.querySelector("canvas");
if (!(canvas instanceof HTMLCanvasElement)) {
throw new Error("Expected a canvas element");
}
const context = canvas.getContext("2d");
if (context === null) {
throw new Error("Canvas 2D is unavailable");
}
const cv = await initOpenCv();
const source = context.getImageData(0, 0, canvas.width, canvas.height);
const result = cv.grayscale(rgbaImageFromImageData(source));
context.putImageData(imageDataFromRgbaImage(result), 0, 0);initOpenCv() loads the WASM module once per call. Keep the returned client and reuse it. Inputs and outputs use copied Uint8Array buffers in 0.1.0. A shared-memory API belongs to a later performance milestone.
Rust-owned matrices are available for pipelines that should stay in WASM memory:
const matrix = cv.matFromU8(1080, 1920, 4, rgbaBytes);
const crop = matrix.roi(100, 200, 480, 640);
try {
const compactCrop = crop.toUint8Array();
// Pass compactCrop to canvas, WebCodecs, or another browser interface.
} finally {
crop.dispose();
matrix.dispose();
}Regions share Rust storage without copying pixels. The core arithmetic, bitwise, comparison, range, and reduction slices accept Mat directly; the original RGBA convenience operations remain available.
Current API
cv.grayscale(image)converts RGB channels to fixed-point BT.601 luma and keeps alpha.cv.cvtColor(source, destination, code, dstCn?)supports U8 color codes 0 through 11 for grayscale, RGB/BGR channel order, and alpha insertion or removal.cv.resize(source, destination, size, fx?, fy?, interpolation?)provides all-depth nearest-neighbor resize plus U8 linear and area modes.cv.invert(image)inverts RGB channels and keeps alpha.cv.threshold(image, value)emits black or white pixels using an inclusive threshold from 0 through 255.cv.resizeNearest(image, width, height)resizes with nearest-neighbor sampling.
Read the API reference for input contracts and conversion helpers.
Parity
The independent browser inventory contains 488 callable families, so the 25% milestone is 122 fully compatible families. A family earns full credit only after every selected browser overload, supported matrix form, output mutation, error case, and differential fixture passes. A useful U8 specialization is recorded as partial and earns no full-parity credit.
| Module | Package method | OpenCV.js family | Status | Current scope |
| ---------- | ------------------------------------------- | ---------------------------------------------- | ------- | ------------------------------------------ |
| core | absdiff | cv.absdiff | Partial | Matching U8 matrices |
| core | add | cv.add | Partial | Saturating U8 matrix operands |
| core | addWeighted | cv.addWeighted | Full | Exact all-depth mutable-output contract |
| core | bitwiseAnd | cv.bitwise_and | Partial | U8 matrix operands, no mask |
| core | bitwiseNot | cv.bitwise_not | Full | All-depth destination/mask contract |
| core | bitwiseOr | cv.bitwise_or | Partial | U8 matrix operands, no mask |
| core | bitwiseXor | cv.bitwise_xor | Partial | U8 matrix operands, no mask |
| core | compareEqual | cv.compare | Partial | U8 equality mode |
| core | countNonZero | cv.countNonZero | Full | Exact all-depth single-channel reduction |
| core | convertScaleAbs | cv.convertScaleAbs | Full | Exact all-depth mutable-output contract |
| core | copyMakeBorder | cv.copyMakeBorder | Partial | All depths and five border modes |
| core | determinant | cv.determinant | Full | Exact F32/F64 square-matrix contract |
| core | divide | cv.divide | Full | Exact all-depth mutable-output contract |
| core | cartToPolar | cv.cartToPolar | Full | Exact paired F32/F64 output contract |
| core | exp | cv.exp | Full | Exact F32/F64 mutable-output contract |
| core | flip | cv.flip | Full | Exact all-depth mutable-output contract |
| core | getLogLevel | cv.getLogLevel | Partial | Package-owned log level 0 through 6 |
| core | getOptimalDFTSize | cv.getOptimalDFTSize | Full | Exact i32 contract and smooth result |
| core | hconcat | cv.hconcat | Partial | All depths, two through four inputs |
| core | inRange | cv.inRange | Partial | U8 matrix bounds |
| core | invert | cv.invert | Partial | Square matrices and three methods |
| core | log | cv.log | Full | Exact F32/F64 mutable-output contract |
| core | lut | cv.LUT | Partial | Byte sources and every table depth |
| core | magnitude | cv.magnitude | Full | Exact matching F32/F64 mutable output |
| core | max | cv.max | Partial | U8 matrix operands |
| core | mean | cv.mean | Full | Exact all-depth optional-mask contract |
| core | meanStdDev | cv.meanStdDev | Partial | All depths, masks, and F64 outputs |
| core | merge | cv.merge | Partial | Two through four all-depth inputs |
| core | min | cv.min | Partial | U8 matrix operands |
| core | minMaxLoc | cv.minMaxLoc | Full | Exact all-depth optional-mask contract |
| core | mixChannels | cv.mixChannels | Partial | One source and destination, all depths |
| core | multiply | cv.multiply | Full | Exact all-depth mutable-output contract |
| core | norm | cv.norm | Partial | All depths, masks, and norm modes |
| core | normalize | cv.normalize | Partial | All depths and mutable destinations |
| core | polarToCart | cv.polarToCart | Full | Exact paired F32/F64 output contract |
| core | perspectiveTransform | cv.perspectiveTransform | Partial | F32/F64 2D and 3D vectors |
| core | pow | cv.pow | Full | Exact all-depth valid-power contract |
| core | randn | cv.randn | Partial | All depths and diagonal deviations |
| core | randu | cv.randu | Partial | All depths and per-channel ranges |
| core | repeat | cv.repeat | Full | Exact all-depth mutable-output contract |
| core | reduce | cv.reduce | Partial | Both axes, four modes, all depths |
| core | rotate | cv.rotate | Full | Exact all-depth mutable-output contract |
| core | setIdentity | cv.setIdentity | Full | Exact Scalar conversion and in-place ROI |
| core | setLogLevel | cv.setLogLevel | Partial | Previous-level return and state update |
| core | setRNGSeed | cv.setRNGSeed | Partial | Deterministic package-owned RNG |
| core | solve | cv.solve | Partial | LU, Cholesky, and QR methods |
| core | subtract | cv.subtract | Partial | Saturating U8 matrix operands |
| core | split | cv.split | Partial | All depths and strided regions |
| core | sqrt | cv.sqrt | Full | Exact F32/F64 mutable-output contract |
| core | transpose | cv.transpose | Full | Exact all-depth mutable-output contract |
| core | trace | cv.trace | Full | Exact all-depth four-lane Scalar |
| core | transform | cv.transform | Partial | All depths and F32/F64 coefficients |
| core | vconcat | cv.vconcat | Partial | All depths, two through four inputs |
| features2d | createAKAZE | cv.AKAZE.create | Partial | Configuration handle only; no detection |
| features2d | AKAZE.getDefaultName | cv.AKAZE.getDefaultName | Full | Exact name, arity, and lifecycle |
| features2d | AKAZE.getDescriptorChannels | cv.AKAZE.getDescriptorChannels | Full | Exact signed i32 state and lifecycle |
| features2d | AKAZE.getDescriptorSize | cv.AKAZE.getDescriptorSize | Full | Exact signed i32 state and lifecycle |
| features2d | AKAZE.getDescriptorType | cv.AKAZE.getDescriptorType | Full | Canonical enum identity and lifecycle |
| features2d | AKAZE.getDiffusivity | cv.AKAZE.getDiffusivity | Full | Shared enum identity and lifecycle |
| features2d | AKAZE.getNOctaveLayers | cv.AKAZE.getNOctaveLayers | Full | Exact signed i32 state and lifecycle |
| features2d | AKAZE.getNOctaves | cv.AKAZE.getNOctaves | Full | Exact signed i32 state and lifecycle |
| features2d | AKAZE.getThreshold | cv.AKAZE.getThreshold | Full | Exact F64 state and lifecycle |
| features2d | AKAZE.setDescriptorChannels | cv.AKAZE.setDescriptorChannels | Full | Exact i32 coercion and call contract |
| features2d | AKAZE.setDescriptorSize | cv.AKAZE.setDescriptorSize | Full | Exact i32 coercion and call contract |
| features2d | AKAZE.setDescriptorType | cv.AKAZE.setDescriptorType | Full | Exact structural enum-object coercion |
| features2d | AKAZE.setDiffusivity | cv.AKAZE.setDiffusivity | Full | Exact structural enum-object coercion |
| features2d | AKAZE.setNOctaveLayers | cv.AKAZE.setNOctaveLayers | Full | Exact i32 coercion and call contract |
| features2d | AKAZE.setNOctaves | cv.AKAZE.setNOctaves | Full | Exact i32 coercion and call contract |
| features2d | AKAZE.setThreshold | cv.AKAZE.setThreshold | Full | Exact number coercion and call contract |
| features2d | createKAZE | cv.KAZE.create | Partial | Configuration handle only; no detection |
| features2d | KAZE.getDefaultName | cv.KAZE.getDefaultName | Full | Exact name, arity, and lifecycle |
| features2d | KAZE.getDiffusivity | cv.KAZE.getDiffusivity | Full | Shared enum identity and lifecycle |
| features2d | KAZE.getExtended | cv.KAZE.getExtended | Full | Exact boolean state and lifecycle |
| features2d | KAZE.getNOctaveLayers | cv.KAZE.getNOctaveLayers | Full | Exact signed i32 state and lifecycle |
| features2d | KAZE.getNOctaves | cv.KAZE.getNOctaves | Full | Exact signed i32 state and lifecycle |
| features2d | KAZE.getThreshold | cv.KAZE.getThreshold | Full | Exact F64 state and lifecycle |
| features2d | KAZE.getUpright | cv.KAZE.getUpright | Full | Exact boolean state and lifecycle |
| features2d | KAZE.setDiffusivity | cv.KAZE.setDiffusivity | Full | Exact structural enum-object coercion |
| features2d | KAZE.setExtended | cv.KAZE.setExtended | Full | Exact boolean coercion and call contract |
| features2d | KAZE.setNOctaveLayers | cv.KAZE.setNOctaveLayers | Full | Exact i32 coercion and call contract |
| features2d | KAZE.setNOctaves | cv.KAZE.setNOctaves | Full | Exact i32 coercion and call contract |
| features2d | KAZE.setThreshold | cv.KAZE.setThreshold | Full | Exact number coercion and call contract |
| features2d | KAZE.setUpright | cv.KAZE.setUpright | Full | Exact boolean coercion and call contract |
| features2d | createMSER | cv.MSER.create | Partial | Static factory absent; no region detection |
| features2d | MSER.getDefaultName | cv.MSER.getDefaultName | Full | Exact name, arity, and lifecycle |
| features2d | MSER.getDelta | cv.MSER.getDelta | Full | Exact signed i32 state and lifecycle |
| features2d | MSER.getMaxArea | cv.MSER.getMaxArea | Full | Exact signed i32 state and lifecycle |
| features2d | MSER.getMinArea | cv.MSER.getMinArea | Full | Exact signed i32 state and lifecycle |
| features2d | MSER.getPass2Only | cv.MSER.getPass2Only | Full | Exact boolean state and lifecycle |
| features2d | MSER.setDelta | cv.MSER.setDelta | Full | Exact i32 coercion and call contract |
| features2d | MSER.setMaxArea | cv.MSER.setMaxArea | Full | Exact i32 coercion and call contract |
| features2d | MSER.setMinArea | cv.MSER.setMinArea | Full | Exact i32 coercion and call contract |
| features2d | MSER.setPass2Only | cv.MSER.setPass2Only | Full | Exact boolean coercion and call contract |
| features2d | createORB | cv.ORB.create | Partial | Static factory absent; no detection |
| features2d | ORB.getDefaultName | cv.ORB.getDefaultName | Full | Exact name, arity, and lifecycle |
| features2d | ORB.getFastThreshold | cv.ORB.getFastThreshold | Full | Exact signed i32 state and lifecycle |
| features2d | ORB.setEdgeThreshold | cv.ORB.setEdgeThreshold | Full | Exact i32 coercion and call contract |
| features2d | ORB.setFastThreshold | cv.ORB.setFastThreshold | Full | Exact i32 coercion and call contract |
| features2d | ORB.setFirstLevel | cv.ORB.setFirstLevel | Full | Exact i32 conversion and validation |
| features2d | ORB.setMaxFeatures | cv.ORB.setMaxFeatures | Full | Exact i32 coercion and call contract |
| features2d | ORB.setNLevels | cv.ORB.setNLevels | Full | Exact i32 coercion and call contract |
| features2d | ORB.setPatchSize | cv.ORB.setPatchSize | Full | Exact i32 coercion and call contract |
| features2d | ORB.setScaleFactor | cv.ORB.setScaleFactor | Full | Exact F64 input and F32 stored state |
| features2d | ORB.setScoreType | cv.ORB.setScoreType | Full | Exact structural enum-object coercion |
| features2d | ORB.setWTA_K | cv.ORB.setWTA_K | Full | Exact i32 coercion and call contract |
| features2d | createAgastFeatureDetector | cv.AgastFeatureDetector.create | Partial | Configuration handle only; no detection |
| features2d | AgastFeatureDetector.getDefaultName | cv.AgastFeatureDetector.getDefaultName | Full | Exact name, arity, and lifecycle |
| features2d | AgastFeatureDetector.getNonmaxSuppression | cv.AgastFeatureDetector.getNonmaxSuppression | Full | Exact boolean state and lifecycle |
| features2d | AgastFeatureDetector.getThreshold | cv.AgastFeatureDetector.getThreshold | Full | Exact signed i32 state and lifecycle |
| features2d | AgastFeatureDetector.getType | cv.AgastFeatureDetector.getType | Full | Canonical enum identity and lifecycle |
| features2d | AgastFeatureDetector.setNonmaxSuppression | cv.AgastFeatureDetector.setNonmaxSuppression | Full | Exact boolean coercion and call contract |
| features2d | AgastFeatureDetector.setThreshold | cv.AgastFeatureDetector.setThreshold | Full | Exact i32 coercion and call contract |
| features2d | AgastFeatureDetector.setType | cv.AgastFeatureDetector.setType | Full | Exact structural enum-object coercion |
| features2d | createFastFeatureDetector | cv.FastFeatureDetector.create | Partial | Configuration handle only; no detection |
| features2d | FastFeatureDetector.getDefaultName | cv.FastFeatureDetector.getDefaultName | Full | Exact name, arity, and lifecycle |
| features2d | FastFeatureDetector.getNonmaxSuppression | cv.FastFeatureDetector.getNonmaxSuppression | Full | Exact boolean state and lifecycle |
| features2d | FastFeatureDetector.getThreshold | cv.FastFeatureDetector.getThreshold | Full | Exact signed i32 state and lifecycle |
| features2d | FastFeatureDetector.getType | cv.FastFeatureDetector.getType | Full | Canonical enum identity and lifecycle |
| features2d | FastFeatureDetector.setNonmaxSuppression | cv.FastFeatureDetector.setNonmaxSuppression | Full | Exact boolean coercion and call contract |
| features2d | FastFeatureDetector.setThreshold | cv.FastFeatureDetector.setThreshold | Full | Exact i32 coercion and call contract |
| features2d | FastFeatureDetector.setType | cv.FastFeatureDetector.setType | Full | Exact structural enum-object coercion |
| features2d | createGFTTDetector | cv.GFTTDetector.create | Partial | Selected six-argument configuration |
| features2d | GFTTDetector.getBlockSize | cv.GFTTDetector.getBlockSize | Full | Exact arity, state, and lifecycle |
| features2d | GFTTDetector.getDefaultName | cv.GFTTDetector.getDefaultName | Full | Exact name, arity, and lifecycle |
| features2d | GFTTDetector.getHarrisDetector | cv.GFTTDetector.getHarrisDetector | Full | Exact boolean state and lifecycle |
| features2d | GFTTDetector.getK | cv.GFTTDetector.getK | Full | Exact F64 state and lifecycle |
| features2d | GFTTDetector.getMaxFeatures | cv.GFTTDetector.getMaxFeatures | Full | Exact signed i32 state and lifecycle |
| features2d | GFTTDetector.getMinDistance | cv.GFTTDetector.getMinDistance | Full | Exact F64 state and lifecycle |
| features2d | GFTTDetector.getQualityLevel | cv.GFTTDetector.getQualityLevel | Full | Exact F64 state and lifecycle |
| features2d | GFTTDetector.setBlockSize | cv.GFTTDetector.setBlockSize | Full | Exact i32 coercion and call contract |
| features2d | GFTTDetector.setHarrisDetector | cv.GFTTDetector.setHarrisDetector | Full | Exact boolean coercion and call contract |
| features2d | GFTTDetector.setK | cv.GFTTDetector.setK | Full | Exact number coercion and call contract |
| features2d | GFTTDetector.setMaxFeatures | cv.GFTTDetector.setMaxFeatures | Full | Exact i32 coercion and call contract |
| features2d | GFTTDetector.setMinDistance | cv.GFTTDetector.setMinDistance | Full | Exact number coercion and call contract |
| features2d | GFTTDetector.setQualityLevel | cv.GFTTDetector.setQualityLevel | Full | Exact number coercion and call contract |
| imgproc | arcLength | cv.arcLength | Full | Exact I32/F32 contour contract |
| imgproc | boundingRect | cv.boundingRect | Full | Exact I32/F32 contour bounds |
| imgproc | clipLine | cv.clipLine | Partial | Integer rectangle and segment form |
| imgproc | contourArea | cv.contourArea | Full | Exact I32/F32 contour area |
| imgproc | createHanningWindow | cv.createHanningWindow | Full | Exact F32/F64 mutable window contract |
| imgproc | ellipse2Poly | cv.ellipse2Poly | Partial | Ordered integer ellipse arcs |
| imgproc | getAffineTransform | cv.getAffineTransform | Full | Exact continuous F32 points and F64 map |
| imgproc | getPerspectiveTransform | cv.getPerspectiveTransform | Partial | Four F32/F64 point pairs to F64 |
| imgproc | getRotationMatrix2D | cv.getRotationMatrix2D | Full | Exact Point2f and F64 matrix contract |
| imgproc | getStructuringElement | cv.getStructuringElement | Full | Exact rectangle, cross, ellipse, diamond |
| imgproc | cvtColor | cv.cvtColor | Partial | U8 RGB/BGR/alpha/gray codes 0 through 11 |
| imgproc | invertAffineTransform | cv.invertAffineTransform | Full | Exact F32/F64 mutable inverse contract |
| imgproc | isContourConvex | cv.isContourConvex | Full | Exact strict-convexity contract |
| imgproc | pointPolygonTest | cv.pointPolygonTest | Full | Exact classification and signed distance |
| imgproc | resize | cv.resize | Partial | Nearest all-depth; U8 linear/area |
| imgproc | threshold | cv.threshold | Partial | Five U8 modes plus Otsu selection |
| imgproc | GaussianBlur | cv.GaussianBlur | Partial | Separable odd U8 kernels and borders |
| imgproc | morphologyEx | cv.morphologyEx | Partial | Seven U8 morphology operations |
| imgproc | Sobel | cv.Sobel | Partial | 3x3 U8 gradients to signed/float outputs |
| imgproc | Canny | cv.Canny | Partial | U8 3x3 gradients, suppression, hysteresis |
| imgproc | findContours | cv.findContours | Partial | External/list U8 contours and MatVector |
| imgproc | warpAffine | cv.warpAffine | Partial | U8 nearest/linear affine image warps |
| imgproc | equalizeHist | cv.equalizeHist | Partial | Exact single-channel U8 equalization |
| imgproc | warpPerspective | cv.warpPerspective | Planned | Not started |
| photo | createTonemapDrago | cv.createTonemapDrago | Partial | Global factory absent; no pixel process |
| photo | createTonemapMantiuk | cv.createTonemapMantiuk | Partial | Global factory absent; no pixel process |
| photo | createTonemapReinhard | cv.createTonemapReinhard | Partial | Global factory absent; no pixel process |
| photo | Tonemap.getGamma | cv.Tonemap.getGamma | Full | Exact inherited F32 state and lifecycle |
| photo | Tonemap.setGamma | cv.Tonemap.setGamma | Full | Exact float coercion and call contract |
| photo | TonemapDrago.getBias | cv.TonemapDrago.getBias | Full | Exact F32 state and lifecycle |
| photo | TonemapDrago.getSaturation | cv.TonemapDrago.getSaturation | Full | Exact F32 state and lifecycle |
| photo | TonemapDrago.setBias | cv.TonemapDrago.setBias | Full | Exact float coercion and call contract |
| photo | TonemapDrago.setSaturation | cv.TonemapDrago.setSaturation | Full | Exact float coercion and call contract |
| photo | TonemapMantiuk.getSaturation | cv.TonemapMantiuk.getSaturation | Full | Exact F32 state and lifecycle |
| photo | TonemapMantiuk.getScale | cv.TonemapMantiuk.getScale | Full | Exact F32 state and lifecycle |
| photo | TonemapMantiuk.setSaturation | cv.TonemapMantiuk.setSaturation | Full | Exact float coercion and call contract |
| photo | TonemapMantiuk.setScale | cv.TonemapMantiuk.setScale | Full | Exact float coercion and call contract |
| photo | TonemapReinhard.getColorAdaptation | cv.TonemapReinhard.getColorAdaptation | Full | Exact F32 state and lifecycle |
| photo | TonemapReinhard.getIntensity | cv.TonemapReinhard.getIntensity | Full | Exact F32 state and lifecycle |
| photo | TonemapReinhard.getLightAdaptation | cv.TonemapReinhard.getLightAdaptation | Full | Exact F32 state and lifecycle |
| photo | TonemapReinhard.setColorAdaptation | cv.TonemapReinhard.setColorAdaptation | Full | Exact float coercion and call contract |
| photo | TonemapReinhard.setIntensity | cv.TonemapReinhard.setIntensity | Full | Exact float coercion and call contract |
| photo | TonemapReinhard.setLightAdaptation | cv.TonemapReinhard.setLightAdaptation | Full | Exact float coercion and call contract |
Current full parity is 124 of 488 (25.41%). There are 53 partial families, for 177 supported families in total. The 25% milestone is complete. bun run parity:check verifies these numbers against the inventory, TypeScript metadata, Rust exports, README rows, and generated JSON.
The fixture passes the complete pinned browser contract for determinant. The function requires exactly one live Mat and accepts only nonempty square single-channel F32 or F64 matrices, including non-contiguous regions. It preserves the input, matches the direct 1x1, 2x2, and 3x3 paths with signed-zero and non-finite propagation, and keeps F32 and F64 arithmetic distinct during elimination for larger matrices. The audit locks the absolute pivot cutoffs, exact cutoff acceptance, row-swap signs, singular positive zero, stored-F32 widening in the small formulas, and Hilbert precision. Integer, multichannel, nonsquare, empty, deleted, and non-Mat inputs reject before computation.
The fixture passes the complete pinned browser contract for exp, log, sqrt, pow, and magnitude. It checks exact arity and Mat conversion, Embind F64 power conversion, typed and canonical empty matrices, destination replacement, detached and shared regions, live overlapping traversal, all valid scalar depths, integer saturation and wrapping, and floating-point bit patterns. The package rejects native calls that expose unsafe uninitialized output while preserving the observable rejection and unchanged-state contract.
The fixture also passes the pinned cartToPolar and polarToCart contracts. It covers both overloads, JavaScript truthiness, F32/F64 multichannel matrices, paired destination replacement, typed empty layouts, full-rotation angles, live overlapping regions, shared-output rejection, and the reference runtime's F32 precision path for both accepted depths.
The fixture passes the complete pinned contracts for multiply, divide, addWeighted, and convertScaleAbs. It covers exact overload dispatch, Embind scalar coercion, all seven depths, explicit mixed-depth conversion, OutputArray replacement, live shared regions, typed empty layouts, floating-point edges, and the reference runtime's non-saturating CV_32S overflow paths.
The fixture passes the complete pinned contracts for arcLength, contourArea, and boundingRect. It covers exact arity and runtime length, JavaScript truthiness, I32 and F32 contours in Nx1C2, 1xNC2, and Nx2C1 layouts, deleted inputs, canonical empty bounds, and rejection of F64, U8, and invalid shapes. The package rejects typed empty contours before entering upstream paths that do not return a safe JavaScript error.
The fixture also passes the complete pinned contracts for isContourConvex and pointPolygonTest. It covers exact arity, strict convexity, continuous I32/F32 layouts, structural Point2f conversion, float32 narrowing, JavaScript truthiness, one-point and segment contours, classification, signed distance, traversal-dependent signed zero, non-finite query sentinels, deleted and empty inputs, and rejected depths, shapes, and regions.
The fixture passes the complete pinned getRotationMatrix2D contract. It checks the exact three-argument call, structural Point2f field order and float32 narrowing, strict Embind double conversion for angle and scale, boolean inputs, signed zero, non-finite propagation, bit-exact 2x3 F64 coefficients, and independent result ownership.
The fixture passes the complete pinned browser contract for rotate, including exact arity and constants, Embind signed i32 conversion, all scalar depths, empty and deleted matrices, OutputArray replacement, in-place operation, detached regions, and live shared-region composition. Invalid native codes preserve the observable no-throw contract without exposing the official build's unsafe output state.
The fixture passes the complete pinned browser contract for repeat, including exact arity, Embind signed i32 conversion, all scalar depths, empty and deleted matrices, OutputArray replacement, compatible and detached regions, exact in-place rejection, and live overlapping shared-region traversal. Invalid native calls are compared by rejection and unchanged state because the official build exposes transient numeric exception pointers; this package preserves stable Rust errors.
The fixture passes the complete pinned browser contract for countNonZero, including exact arity and Mat conversion errors, all seven scalar depths available in the artifact, signed zero, NaN, infinities, subnormal values, fresh empty matrices, deleted handles, and non-contiguous regions. Multi-channel rejection is compared semantically because the official build throws a transient numeric native-exception pointer; this package preserves a stable Rust error instead.
The fixture passes the complete pinned browser contract for bitwiseNot. It covers exact overload dispatch, all scalar depths and one through five channels, raw floating-point bit patterns, optional U8/I8 masks, destination reuse and replacement, typed empties, strided regions, exact aliases, live unmasked overlaps, masked overlaps, argument conversion order, invalid inputs, and deleted handles.
The fixture passes the complete pinned browser contracts for mean and minMaxLoc. It covers both optional-mask overloads, all scalar depths, one through four mean channels, single-channel extrema, compact and strided matrices, malformed masks, empty-header sentinels, first row-major ties, NaN, infinities, signed zero, and deleted or invalid inputs.
The fixture passes the complete pinned browser contract for trace. It checks exact arity, all scalar depths, one through four diagonal channel sums, rectangular and strided matrices, typed empties, F32 widening, F64 accumulation order, signed zero, NaN, infinities, and invalid or deleted inputs.
The fixture passes the complete pinned browser contract for getOptimalDFTSize, including exact arity, Embind signed i32 coercion and errors, negative and zero inputs, every smooth-size boundary exercised by Rust tests, and the exclusive 2,125,764,000 upper sentinel. This family counts as full parity.
The fixture passes the complete pinned browser contract for all 15 AKAZE instance methods. In addition to the scalar checks, the four enum methods match the official enum namespaces, canonical singleton identity, structural setter coercion, raw unknown wire values, arity errors, and lifetime behavior. The factory remains partial because the pinned artifact omits its config-listed static constructor.
The fixture passes the complete pinned browser contract for all 13 KAZE instance methods. Its diffusivity getter returns the shared canonical enum singleton, and its setter applies the official structural enum-object conversion while preserving unknown signed i32 wire values. The factory remains partial because the pinned artifact omits its config-listed static constructor.
The fixture passes the complete pinned browser contract for all seven AGAST and all seven FAST instance methods. The type methods match the official enum namespaces, canonical singleton identity, structural setter coercion, raw unknown wire values, arity errors, and lifetime behavior. Both factories remain partial because the pinned artifact omits their config-listed static constructors.
The fixture passes the complete pinned browser matrix for all 13 GFTT instance methods. It checks method arity, defaults, return values, integer, number, and boolean coercion, missing and extra arguments, deletion, repeat deletion, and calls after deletion. These 13 method families count as full parity. The official artifact exposes the direct constructor but omits the config-listed static GFTTDetector.create method. The package factory remains partial because it covers one six-argument shape and omits the gradientSize overload.
Read the inventory and complete parity contract for the denominator, exclusions, and definition of done.
What we build next
The Rust Mat owns U8, I8, U16, I16, I32, F32, and F64 storage, plus channels, dimensions, byte strides, zero-copy regions, mutable destinations, WASM allocation, and deterministic disposal. The next foundation broadens arithmetic and image-processing families across every scalar depth while differential fixtures lock behavior to the pinned browser baseline.
The active image-processing slice finishes the remaining color conversions, then adds the resize interpolation modes in the pinned OpenCV.js baseline, convolution, Gaussian blur, Sobel gradients, and Canny. Each operation needs upstream differential fixtures and real-browser benchmarks before its parity status changes to implemented.
The performance goal is at least 2x the OpenCV.js geometric mean for warmed 1080p hot kernels and at least 4x for fused pipelines. Those are targets, not current results. Reports will include p50 and p95 timing, initialization, allocation counts, package bytes, scalar fallback results, and SIMD results. See the performance contract.
Browser and bundler requirements
Consumers need ES2022 modules, WebAssembly, dynamic import(), and support for package imports maps. Modern Vite, Rollup, webpack, and esbuild-based browser builds normally provide these pieces. The package does not include Node-specific image decoding or DOM polyfills.
Develop locally
Install Bun, Rust, and the wasm32-unknown-unknown Rust target. The npm wasm-pack development dependency supplies the WASM build command.
bun install
bun run check
bun run buildGenerated JavaScript and declarations go to dist/. The generated WASM loader and binary go to wasm/. Git ignores both directories; npm includes both in the packed release.
Project documents
- Architecture
- API reference
- OpenCV parity
- OpenCV.js browser inventory
- Performance contract
- Source-independent compatibility policy
- Licensing research
- OpenCV license and implementation review
- Package-name review
- Third-party notices
- Roadmap
- TODO
- Contributing
- Versioning
- Publishing
- Security
- Changelog
License
MIT. See LICENSE.
OpenCV is a trademark of its owner. WASMosaic is an independent project and is not affiliated with or endorsed by OpenCV. The OpenCV name is used only to identify the compatibility target.