JKCS/src/main.ts

import * as tf from '@tensorflow/tfjs';
import { setWasmPaths } from '@tensorflow/tfjs-backend-wasm';
import modelJSON from './model.json';

const charset = [' ', '0', '2', '4', '8', 'A', 'D', 'G', 'H', 'J', 'K', 'M', 'N', 'P', 'R', 'S', 'T', 'V', 'W', 'X', 'Y'];
let weightsData: Uint8Array; // base64 encoded weights
let model: tf.LayersModel;

tf.enableProdMode();

const wasmToUrl = (wasm: any) => {
  const blb = new Blob([wasm], { type: 'application/wasm' });
  return URL.createObjectURL(blb);
};

const backendloaded = (async () => {
  try {
    // dead code elimination should occur here
    // eslint-disable-next-line camelcase
    if (execution_mode === 'userscript' || execution_mode === 'test') {
      weightsData = (await import('./group1-shard1of1.bin')).default;
      const tfwasmthreadedsimd = (await import('./tfjs-backend-wasm-threaded-simd.wasm')).default;
      const tfwasmsimd = (await import('./tfjs-backend-wasm-simd.wasm')).default;
      const tfwasm = (await import('./tfjs-backend-wasm.wasm')).default;
      setWasmPaths({
        'tfjs-backend-wasm.wasm': wasmToUrl(tfwasm),
        'tfjs-backend-wasm-simd.wasm': wasmToUrl(tfwasmsimd),
        'tfjs-backend-wasm-threaded-simd.wasm': wasmToUrl(tfwasmthreadedsimd)
      });
    } else {
      weightsData = new Uint8Array(await (await fetch(chrome.runtime.getURL('./group1-shard1of1.bin'))).arrayBuffer());
      const args = {
        'tfjs-backend-wasm.wasm': chrome.runtime.getURL('tfjs-backend-wasm.wasm'),
        'tfjs-backend-wasm-simd.wasm': chrome.runtime.getURL('tfjs-backend-wasm-simd.wasm'),
        'tfjs-backend-wasm-threaded-simd.wasm': chrome.runtime.getURL('tfjs-backend-wasm-threaded-simd.wasm')
      };
      setWasmPaths(args);
    }
    const l = await tf.setBackend('wasm');
    console.log('tf backend loaded', l);
  } catch (err) {
    console.log('tf err', err);
  }
})();

function toggle(obj: HTMLElement, v?: any) {
  if (v) obj.style.display = '';
  else obj.style.display = 'none';
}

const iohander: tf.io.IOHandler = {
  load: function () {
    return new Promise((resolve, reject) => {
      resolve({
        modelTopology: modelJSON.modelTopology,
        weightSpecs: modelJSON.weightsManifest[0].weights as tf.io.WeightsManifestEntry[],
        weightData: weightsData.buffer,
        format: modelJSON.format,
        generatedBy: modelJSON.generatedBy,
        convertedBy: modelJSON.convertedBy
      });
    });
  }
};

async function load() {
  model = await tf.loadLayersModel(iohander);
  return model;
}

function black(x: number) {
  return x < 64;
}

// Calculates "disorder" of the image. "Disorder" is the percentage of black pixels that have a
// non-black pixel below them. Minimizing this seems to be good enough metric for solving the slider.
function calculateDisorder(imgdata: ImageData) {
  const a = imgdata.data;
  const w = imgdata.width;
  const h = imgdata.height;
  const pic: number[] = [];
  const visited: number[] = [];

  for (let c = 0; c < w * h; c++) {
    if (visited[c]) continue;
    if (!black(a[c * 4])) continue;

    let blackCount = 0;
    const items: number[] = [];
    const toVisit: number[] = [c];
    while (toVisit.length > 0) {
      const cc = toVisit[toVisit.length - 1];
      toVisit.splice(toVisit.length - 1, 1);

      if (visited[cc]) continue;
      visited[cc] = 1;

      if (black(a[cc * 4])) {
        items.push(cc);

        blackCount++;
        toVisit.push(cc + 1);
        toVisit.push(cc - 1);
        toVisit.push(cc + w);
        toVisit.push(cc - w);
      }
    }

    if (blackCount >= 24) {
      items.forEach(function (x) {
        pic[x] = 1;
      });
    }
  }

  let res = 0;
  let total = 0;
  for (let c = 0; c < w * h - w; c++) {
    if (pic[c] !== pic[c + w]) res += 1;
    if (pic[c]) total += 1;
  }

  return res / (total === 0 ? 1 : total);
}

/*
 * decide if a pixel is closer to black than to white.
 * return 0 for white, 1 for black
 */
function pxlBlackOrWhite(r: number, g: number, b: number) {
  return (r + g + b > 384) ? 0 : 1;
}

/*
 * Get bordering pixels of transparent areas (the outline of the circles)
 * and return their coordinates with the neighboring color.
 */
function getBoundries(imgdata: ImageData) {
  const data = imgdata.data;
  const width = imgdata.width;

  let i = data.length - 1;
  let cl = 0;
  let cr = 0;
  const chkArray = [];
  let opq = true;
  while (i > 0) {
    // alpha channel above 128 is assumed opaque
    const a = data[i] > 128;
    if (a !== opq) {
      if ((data[i - 4] > 128) === opq) {
        // ignore just 1-width areas
        i -= 4;
        continue;
      }
      if (a) {
        /* transparent pixel to its right */
        /*
                // set to color blue (for debugging)
                data[i + 4] = 255;
                data[i + 3] = 255;
                data[i + 2] = 0;
                data[i + 1] = 0;
                */
        const pos = (i + 1) / 4;
        const x = pos % width;
        const y = (pos - x) / width;
        // 1: black, 0: white
        const clr = pxlBlackOrWhite(data[i - 1], data[i - 2], data[i - 3]);
        chkArray.push([x, y, clr]);
        cr += 1;
      } else {
        /* opaque pixel to its right */
        /*
                // set to color red (for debugging)
                data[i] = 255;
                data[i - 1] = 0;
                data[i - 2] = 0;
                data[i - 3] = 255;
                */
        const pos = (i - 3) / 4;
        const x = pos % width;
        const y = (pos - x) / width;
        // 1: black, 0: white
        const clr = pxlBlackOrWhite(data[i + 1], data[i + 2], data[i + 3]);
        chkArray.push([x, y, clr]);
        cl += 1;
      }
      opq = a;
    }
    i -= 4;
  }
  return chkArray;
}
/*
 * slide the background image and compare the colors of the border pixels in
 * chkArray, the position with the most matches wins
 * Return in slider-percentage.
 */
function getBestPos(bgdata: ImageData, chkArray: number[][], slideWidth: number) {
  const data = bgdata.data;
  const width = bgdata.width;
  let bestSimilarity = 0;
  let bestPos = 0;

  for (let s = 0; s <= slideWidth; s += 1) {
    let similarity = 0;
    const amount = chkArray.length;
    for (let p = 0; p < amount; p += 1) {
      const chk = chkArray[p];
      const x = chk[0] + s;
      const y = chk[1];
      const clr = chk[2];
      const off = (y * width + x) * 4;
      const bgclr = pxlBlackOrWhite(data[off], data[off + 1], data[off + 2]);
      if (bgclr === clr) {
        similarity += 1;
      }
    }
    if (similarity > bestSimilarity) {
      bestSimilarity = similarity;
      bestPos = s;
    }
  }
  return bestPos / slideWidth * 100;
}

async function getImageDataFromURI(uri: string) {
  const image = await imageFromUri(uri);
  if (!image)
    throw new Error("No image");
  const canvas = document.createElement('canvas');
  canvas.width = image.width;
  canvas.height = image.height;
  const ctx = canvas.getContext('2d')!;
  ctx.drawImage(image, 0, 0);
  return ctx.getImageData(0, 0, canvas.width, canvas.height);
}

async function slideCaptcha(tfgElement: HTMLElement, tbgElement: HTMLElement, sliderElement: HTMLInputElement) {
  // get data uris for captcha back- and foreground
  const tbgUri = tbgElement.style.backgroundImage.slice(5, -2);
  const tfgUri = tfgElement.style.backgroundImage.slice(5, -2);

  // load foreground (image with holes)
  const igd = await getImageDataFromURI(tfgUri);
  // get array with pixels of foreground
  // that we compare to background
  const chkArray = getBoundries(igd);
  // load background (image that gets slid)
  const sigd = await getImageDataFromURI(tbgUri);
  const slideWidth = sigd.width - igd.width;
  // slide, compare and get best matching position
  const sliderPos = getBestPos(sigd, chkArray, slideWidth);
  // slide in the UI
  sliderElement.value = '' + sliderPos;
  (sliderElement as any).dispatchEvent(new Event('input'), { bubbles: true });
  return 0 - (sliderPos / 2);
}

// returns ImageData from captcha's background image, foreground image, and offset (ranging from 0 to -50)
async function imageFromCanvas(img: HTMLImageElement, bg: HTMLImageElement, off: number | null) {
  const h = img.height;
  const w = img.width;
  const th = 80;
  const ph = 0;
  const pw = 16;
  const scale = th / h;

  const canvas = document.createElement('canvas');
  const fcanvas = document.createElement('canvas');
  const cw = w * scale + pw * 2;
  canvas.width = cw >= 300 ? 300 : cw;
  canvas.height = th;

  fcanvas.width = 300;
  fcanvas.height = 80;

  const ctx = canvas.getContext('2d')!;
  const fctx = fcanvas.getContext('2d')!; // used to contain the captcha stretched to 300w

  ctx.fillStyle = 'rgb(238,238,238)';
  ctx.fillRect(0, 0, canvas.width, canvas.height);

  ctx.translate(canvas.width / 2, canvas.height / 2);

  const draw = function (off: number) {
    if (bg) {
      const border = 4;
      ctx.drawImage(
        bg,
        -off + border,
        0,
        w - border * 2,
        h,
        -w / 2 + border,
        -h / 2,
        w - border * 2,
        h
      );
    }
    ctx.drawImage(img, -w / 2, -h / 2, w, h);
    fctx.drawImage(canvas, 0, 0, 300, 80);
  };

  if (bg && off == null) {
    off = await slideCaptcha(document.getElementById('t-fg')!, document.getElementById('t-bg')!, document.getElementById('t-slider') as HTMLInputElement);
  }
  draw(off || 0);
  return fctx.getImageData(0, 0, 300, 80);
}

function toMonochromeFloat(px: Uint8ClampedArray) {
  const ret = Array<number>(px.length >> 2);
  for (let i = 0; i < px.length; i += 4) {
    ret[i >> 2] = px[i] / 255;
  }
  return ret;
}

const greedyCTCDecode = (yPred: tf.Tensor<tf.Rank>) => tf.tidy(() => yPred.argMax(-1).arraySync());

function processCTCDecodedSequence(decodedSequence: number[], blankLabel = 0) {
  const result = [];
  let prevLabel = blankLabel;

  for (const label of decodedSequence) {
    if (label !== blankLabel && label !== prevLabel) {
      result.push(label);
    }
    prevLabel = label;
  }

  return result;
}

function indicesToSymbols(decodedIndices: number[]) {
  return decodedIndices.map(index => charset[index - 1] || '');
}

async function predict(img: HTMLImageElement, bg: HTMLImageElement, off: number) {
  if (!model) {
    model = await load();
  }
  const image = await imageFromCanvas(img, bg, off);
  if (!image)
    throw new Error("Failed to gen image");
  const mono = toMonochromeFloat(image.data);

  const filtered2 = tf.tensor3d(mono, [image.height, image.width, 1]);
  const prediction = model.predict(filtered2.transpose([1, 0, 2]).expandDims(0));

  let d: tf.TypedArray;

  if (Array.isArray(prediction))
    throw new Error("Unexpected inference type");

  const v = greedyCTCDecode(prediction) as number[][];
  const s = processCTCDecodedSequence(v[0], charset.length + 1);
  return indicesToSymbols(s).join('').trim();
}

async function imageFromUri(uri: string) {
  if (uri.startsWith('url("')) {
    uri = uri.substr(5, uri.length - 7);
  }
  // eslint-disable-next-line camelcase
  if (execution_mode !== 'test' && !uri.startsWith('data:')) {
    return null;
  }

  const img = new Image();
  await new Promise((r) => {
    img.onload = r;
    img.src = uri;
  });
  return img;
}

async function predictUri(uri: string, uribg: string, bgoff: string | null) {
  const img = await imageFromUri(uri);
  const bg = uribg ? await imageFromUri(uribg) : null;
  const off = bgoff ? parseInt(bgoff) : null;

  return await predict(img!, bg!, off!);
}

const solveButton = document.createElement('input');
solveButton.id = 't-auto-solve';
solveButton.value = 'Solve';
solveButton.type = 'button';
solveButton.style.fontSize = '11px';
solveButton.style.padding = '0 2px';
solveButton.style.margin = '0px 0px 0px 6px';
solveButton.style.height = '18px';
solveButton.onclick = async function () {
  solve(true);
};

const altsDiv = document.createElement('div');
altsDiv.id = 't-auto-options';
altsDiv.style.margin = '0';
altsDiv.style.padding = '0';

let storedPalceholder: string;

let overrides = {};

function placeAfter(elem: HTMLElement, sibling: HTMLElement) {
  if (elem.parentElement !== sibling.parentElement) {
    setTimeout(function () {
      sibling.parentElement?.insertBefore(elem, sibling.nextElementSibling);
    }, 1);
  }
}

let previousText: string | null = null;
async function solve(force?: any) {
  const resp = document.getElementById('t-resp') as HTMLInputElement;
  if (!resp) return;

  const bg = document.getElementById('t-bg');
  if (!bg) return;

  const fg = document.getElementById('t-fg');
  if (!fg) return;

  const help = document.getElementById('t-help');
  if (!help) return;

  await backendloaded;

  placeAfter(solveButton, resp);
  placeAfter(altsDiv, help);

  // palememe
  setTimeout(function () {
    toggle(solveButton, bg.style.backgroundImage);
  }, 1);

  const text = fg.style.backgroundImage;
  if (!text) {
    altsDiv.innerHTML = '';
    return;
  }

  if (text === previousText && !force) return;
  previousText = text;

  altsDiv.innerHTML = '';
  if (!storedPalceholder)
    storedPalceholder = resp.placeholder;
  resp.placeholder = 'solving captcha...';

  overrides = {};

  const result = await predictUri(
    text,
    bg.style.backgroundImage,
    force ? bg.style.backgroundPositionX : null
  );
  resp.placeholder = storedPalceholder;

  resp.value = result;
}

const observer = new MutationObserver(async function (mutationsList, observer) {
  solve(false);
});

//window['solve'] = solve;

observer.observe(document.body, {
  attributes: true,
  childList: true,
  subtree: true
});