Implementing the Sobel image edge detection algorithm using WebAssembly

I heard about WebAssembly a long time ago and have a general idea that it offers high execution speed, suitable for scenarios with high-performance requirements like gaming, image/video editing, and more.

So this time, I plan to try implementing the Sobel image edge detection algorithm separately using native JavaScript and WebAssembly to compare the performance difference between the two.

Here, I won’t delve into the detailed implementation of the Sobel algorithm, but rather focus on how to implement it using WebAssembly.

Thanks to miguelmota, there is already a ready-made JavaScript version implementation of the Sobel algorithm, what I need to do is translate it into C++ and compile it into WebAssembly.

// sobel.c

#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <emscripten/emscripten.h>
#include <stdint.h>

EMSCRIPTEN_KEEPALIVE
uint8_t *create_buffer(int width, int height)
{
    return malloc(width * height * 4 * sizeof(uint8_t));
}

EMSCRIPTEN_KEEPALIVE
void destroy_buffer(uint8_t *p)
{
    free(p);
}

int result[1];
EMSCRIPTEN_KEEPALIVE
void sobel(uint8_t *img_in, int width, int height)
{
    uint8_t *img_out;
    img_out = (uint8_t *)malloc(width * height * 4 * sizeof(uint8_t));

    const int Gx[3][3] = {
        {-1, 0, 1},
        {-2, 0, 2},
        {-1, 0, 1}};

    const int Gy[3][3] = {
        {-1, -2, -1},
        {0, 0, 0},
        {1, 2, 1}};

    for (int y = 1; y < height - 1; y++)
    {
        for (int x = 1; x < width - 1; x++)
        {
            int sumX = 0;
            int sumY = 0;

            for (int j = 0; j < 3; j++)
            {
                for (int i = 0; i < 3; i++)
                {
                    const int pixelIndex = (y + j - 1) * width + (x + i - 1);
                    const int pixelValue = img_in[pixelIndex * 4];

                    sumX += pixelValue * Gx[j][i];
                    sumY += pixelValue * Gy[j][i];
                }
            }

            int gradientMagnitude = sqrt(sumX * sumX + sumY * sumY);
            const int outputPixelIndex = (y * width + x) * 4;

            // Converting an int to uint8_t may result in overflow
            if (gradientMagnitude > 255)
            {
                gradientMagnitude = 255;
            }

            img_out[outputPixelIndex] = gradientMagnitude;
            img_out[outputPixelIndex + 1] = gradientMagnitude;
            img_out[outputPixelIndex + 2] = gradientMagnitude;
            // Alpha channel
            img_out[outputPixelIndex + 3] = 255;
        }
    }

    result[0] = (int)img_out;
}

EMSCRIPTEN_KEEPALIVE
void free_result(uint8_t *result)
{
    free(result);
}

EMSCRIPTEN_KEEPALIVE
int get_result_pointer()
{
    return result[0];
}

Then, we can compile the C code into WebAssembly. I use Docker to run the command, and you can find a guide on how to install Emscripten here.

docker run --rm -v $(pwd):/src -u $(id -u):$(id -g) \
  emscripten/emsdk emcc sobel.c -O3 -o sobel.js -s EXTRA_EXPORTED_RUNTIME_METHODS='["cwrap"]'

After compilation, we got two files: sobel.js and sobel.wasm, where sobel.js is the glue code, and sobel.wasm is the WebAssembly code.

Then let’s write the html and javascript to test the function.

<!DOCTYPE html>
<html lang="en">
  <head>
    <meta charset="UTF-8" />
    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
    <title>Document</title>
    <style>
      canvas {
        border: 1px solid;
      }
    </style>
  </head>
  <body>
    <div><input type="file" id="input-file" /></div>
    <div>
      <p>Original image</p>
      <canvas id="canvas1"></canvas>
    </div>
    <div>
      <p>WebAssembly sobel</p>
      <canvas id="canvas2"></canvas>
    </div>

    <div>
      <p>Javascript sobel</p>
      <canvas id="canvas3"></canvas>
    </div>

    <script src="./sobel.js"></script>
    <script src="./index.js"></script>
  </body>
</html>

const inputFile = document.getElementById("input-file");
const canvas1 = document.getElementById("canvas1");
const canvas2 = document.getElementById("canvas2");
const canvas3 = document.getElementById("canvas3");

const ctx1 = canvas1.getContext("2d", { willReadFrequently: true });
const ctx2 = canvas2.getContext("2d", { willReadFrequently: true });
const ctx3 = canvas3.getContext("2d", { willReadFrequently: true });

async function loadImage(src) {
  // Load image
  const imgBlob = await fetch(src).then(resp => resp.blob());
  const img = await createImageBitmap(imgBlob);
  // Make canvas same size as image
  canvas1.width = img.width;
  canvas1.height = img.height;

  // Draw image
  ctx1.drawImage(img, 0, 0);
  return ctx1.getImageData(0, 0, img.width, img.height);
}

const Gx = [
  [-1, 0, 1],
  [-2, 0, 2],
  [-1, 0, 1],
];
const Gy = [
  [-1, -2, -1],
  [0, 0, 0],
  [1, 2, 1],
];

// Sobel image edge detection algorithm in Javascript
function sobelAlgorithm(imageData) {
  const width = imageData.width;
  const height = imageData.height;
  const outputData = new Uint8ClampedArray(imageData.data.length);

  for (let y = 1; y < height - 1; y++) {
    for (let x = 1; x < width - 1; x++) {
      let sumX = 0;
      let sumY = 0;

      for (let j = 0; j < 3; j++) {
        for (let i = 0; i < 3; i++) {
          const pixelIndex = (y + j - 1) * width + (x + i - 1);
          const pixelValue = imageData.data[pixelIndex * 4];
          sumX += pixelValue * Gx[j][i];
          sumY += pixelValue * Gy[j][i];
        }
      }

      const gradientMagnitude = Math.sqrt(sumX * sumX + sumY * sumY);
      // Rounding to an integer
      const normalizedMagnitude = gradientMagnitude >>> 0;

      const outputPixelIndex = (y * width + x) * 4;
      outputData[outputPixelIndex] = normalizedMagnitude;
      outputData[outputPixelIndex + 1] = normalizedMagnitude;
      outputData[outputPixelIndex + 2] = normalizedMagnitude;
      outputData[outputPixelIndex + 3] = 255; // Alpha channel
    }
  }

  return new ImageData(outputData, width, height);
}

// Detect image edges using the Sobel algorithm in both native JavaScript and WebAssembly, and draw them to canvas
function appleSobelDrawImageData(api, imageData) {
  const { width, height } = imageData;

  canvas2.width = width;
  canvas2.height = height;
  canvas3.width = width;
  canvas3.height = height;

  const p = api.create_buffer(width, height);
  Module.HEAPU8.set(imageData.data, p);

  console.time("codeExecution c++");
  api.sobel(p, width, height);
  console.timeEnd("codeExecution c++");

  const resultPointer = api.get_result_pointer();
  const resultView = new Uint8ClampedArray(
    Module.HEAPU8.buffer,
    resultPointer,
    width * height * 4
  );
  api.free_result(resultPointer);
  api.destroy_buffer(p);

  const outImageData = new ImageData(resultView, width, height);
  ctx2.putImageData(outImageData, 0, 0);

  console.time("codeExecution javascript");
  const sobelData = sobelAlgorithm(imageData);
  console.timeEnd("codeExecution javascript");

  ctx3.putImageData(sobelData, 0, 0);
}

Module.onRuntimeInitialized = _ => {
  // Create wrapper functions for all the exported C functions
  const api = {
    create_buffer: Module.cwrap("create_buffer", "number", [
      "number",
      "number",
    ]),
    destroy_buffer: Module.cwrap("destroy_buffer", "", ["number"]),
    gray_scale: Module.cwrap("gray_scale", "", ["number", "number", "number"]),
    sobel: Module.cwrap("sobel", "", ["number", "number", "number"]),
    free_result: Module.cwrap("free_result", "", ["number"]),
    get_result_pointer: Module.cwrap("get_result_pointer", "number", []),
  };

  loadImage("bocchi.JPG").then(imageData => {
    appleSobelDrawImageData(api, imageData);
  });

  inputFile.addEventListener("change", event => {
    const file = event.target.files[0];

    if (file) {
      const reader = new FileReader();
      reader.onload = function (e) {
        const img = new Image();

        img.onload = function () {
          const { width, height } = img;
          canvas1.width = width;
          canvas1.height = height;
          ctx1.drawImage(img, 0, 0);
          const imageData = ctx1.getImageData(0, 0, width, height);

          appleSobelDrawImageData(api, imageData);
        };

        img.src = e.target.result;
      };
      reader.readAsDataURL(file);
    }
  });
};

And the result, pretty good. Sobel image edge detection

Performance Comparison between JavaScript and WebAssembly, achieving nearly tenfold speed improvement.

codeExecution c++: 12.4970703125 ms
codeExecution javascript: 100.006103515625 ms

You can find the source code for this example here, and you can try it out online here.The algorithm’s execution time will be logged to the console.

Implementing the Sobel image edge detection algorithm using WebAssembly

Reference Links