// ==UserScript==
// @name 基于tf的zf验证码填充脚本
// @version 1.0
// @author 1248289414
// @namespace https://github.com/1248289414
// @description 基于tensorflow.js的自动填充教务处验证码的脚本
// @match https://jwc.scnu.edu.cn/
// @match https://jwc.scnu.edu.cn/default2.aspx
// @run-at document-end
// @require https://cdn.jsdelivr.net/npm/@tensorflow/tfjs@1.1.2/dist/tf.min.js
// @grant GM_registerMenuCommand
// ==/UserScript==
// 结果跟实际字母的对照表
const Alphabet = "012345678abcdefghijklmnpqrstuvwxy";
const DBname = "jwc-captcha-model-1";
class imgData {
async load(img) {
// 整个图片的像素个数
let IMAGE_SIZE = img.width * img.height;
// 获取canvas用于绘图
let canvas = document.createElement('canvas');
canvas.width = img.width;
canvas.height = img.height;
// 绘图
let ctx = canvas.getContext('2d');
ctx.drawImage(img, 0, 0);
// 将提取像素信息
let imageData = ctx.getImageData(0, 0, canvas.width, canvas.height);
// 单独提取red通道值，建立灰度图
let grayImageData = new Float32Array(IMAGE_SIZE);
for (let i = 0; i < IMAGE_SIZE; i++) {
grayImageData[i] = imageData.data[i * 4] / 255.0;
}
// 在cut_x的切割点处将，切割图像为4个18*23的小块
let cut_x = [0, 13, 27, 41];
let block_width = 18;
let block_height = 23;
let block_size = block_width * block_height;
this.data = new Float32Array(block_size * 4);
for (let i = 0; i < 4; i++) {
let index = cut_x[i];
for (let j = 0; j < block_height; j++) {
let row = grayImageData.slice(j * img.width + index, j * img.width + index + block_width);
this.data.set(row, i * block_size + j * block_width)
}
}
}
get() {
return tf.tensor4d(this.data, [4, 23, 18, 1]);
}
}
async function run() {
// 获取验证码图片，判断是否加载成功
let img = document.getElementById("icode");
if (img.height != 27 || img.width != 72) {
return
}
let data = new imgData();
await data.load(img);
await indexedDB.open(DBname);
// 打开模型，失败则退出
let model;
try {
model = await tf.loadLayersModel("indexeddb://" + DBname);
} catch (e) {
console.log("模型加载失败，你可能没有导入模型");
return
}
// 预测验证码的值
let score = await model.predict(data.get()).argMax([-1]);
// 将结果转换为字符串
let arr = Array.from(score.dataSync());
let captcha = '';
for (let i = 0; i < 4; i++) {
captcha = captcha + Alphabet[arr[i]];
}
// 填入验证码
let txtSecretCode = document.getElementById("txtSecretCode");
txtSecretCode.value = captcha;
}
// 上传模型并保存
async function get_and_storage_model() {
let json_upload = document.createElement("input");
json_upload.id = "json_upload";
json_upload.type = "file";
let weights_upload = document.createElement("input");
weights_upload.id = "weights_upload";
weights_upload.type = "file";
json_upload.onchange = async function () {
confirm("传入权重文件，后缀名为.bin")
weights_upload.click()
}
weights_upload.onchange = async function () {
if (json_upload.files.length <= 0 && weights_upload.files.length <= 0) {
confirm("你没有选择文件");
return
}
await indexedDB.open(DBname);
let model = await tf.loadLayersModel(tf.io.browserFiles([json_upload.files[0], weights_upload.files[0]]));
await model.save("indexeddb://" + DBname);
run();
}
confirm("传入模型文件，后缀名为.json")
json_upload.click()
}
(async function () {
'use strict';
GM_registerMenuCommand("导入模型", get_and_storage_model);
// 运行预测
run();
let img = document.getElementById("icode");
img.onload = run;
})();
