添加了新类型验证码 dots_and_chars

dots_and_chars's score: 96.0% (48/50)
使用 sharp 和纯 JavaScript 的 CV 算法，不依赖OpenCV

README.md

@@ -3,10 +3,10 @@
 
 <p align="center"><b>使用 CV (OpenCV) 和 OCR (Tesseract) 进行验证码识别</b></p>
 
-simplest | grids_and_equations | ...
-:-: | :-: | :-:
-<img src="./docs/img/simplest.jpg" height="20" alt="simplest" align=center> | <img src="./docs/img/grids_and_equations.jpg" height="20" alt="grids_and_equations" align=center> | ...
-2348 | 2x6=? | ...
+simplest | grids_and_equations | dots_and_chars | ...
+:-: | :-: | :-: | :-:
+<img src="./docs/img/simplest.jpg" height="20" alt="simplest" align=center> | <img src="./docs/img/grids_and_equations.jpg" height="20" alt="grids_and_equations" align=center> | <img src="./docs/img/dots_and_chars.gif" height="20" alt="dots_and_chars" align=center> | ...
+2348 | 2x6=? | 7RVO | ...
 
 ## 快速入门
 
@@ -17,6 +17,8 @@ simplest | grids_and_equations | ...
 npm i opencv4nodejs -g
 ```
 
+第三种验证码识别改为用 [sharp](https://github.com/lovell/sharp) 和纯 JavaScript 的 CV 算法来实现，方便在树莓派上运行，但效率相比前两者很低。
+
 第二个 Tesseract 支持模块为 [tesseract.js](https://github.com/naptha/tesseract.js) 
 
 直接安装
@@ -58,13 +60,16 @@ var mode = "simplest";
 ## 开发
 
 ### 已支持
-simplest | grids_and_equations
-:-: | :-:
-<img src="./docs/img/simplest.jpg" height="20" alt="simplest" align=center> | <img src="./docs/img/grids_and_equations.jpg" height="20" alt="grids_and_equations" align=center>
+simplest | grids_and_equations | dots_and_chars
+:-: | :-: | :-:
+<img src="./docs/img/simplest.jpg" height="20" alt="simplest" align=center> | <img src="./docs/img/grids_and_equations.jpg" height="20" alt="grids_and_equations" align=center> | <img src="./docs/img/dots_and_chars.gif" height="20" alt="dots_and_chars" align=center>
+2348 | 2x6=? | 7RVO
 
 ### 新支持
 
-codes下的文件夹对应着不同种类的名字（自行命名），你可以参照已有的模板与API创建新的识别库，来适配其他各种验证码。请珍惜 [opencv4nodejs](https://github.com/justadudewhohacks/opencv4nodejs) 和 [tesseract.js](https://github.com/naptha/tesseract.js) 的文档:
+codes下的文件夹对应着不同种类的名字（自行命名），你可以参照已有的模板与API创建新的识别库，来适配其他各种验证码。
+
+### 参考文档与额外说明
 
 - opencv4nodejs
     - Github | https://github.com/justadudewhohacks/opencv4nodejs
@@ -73,18 +78,12 @@ codes下的文件夹对应着不同种类的名字（自行命名），你可以
     - 主页 | https://tesseract.projectnaptha.com/
     - Github | https://github.com/naptha/tesseract.js
     - API | https://github.com/naptha/tesseract.js#docs
-
-另外 C++ / Python 的 OpenCV 海量资料也非常有帮助， 相应的函数基本都能在 [opencv4nodejs 的 API 文档](https://justadudewhohacks.github.io/opencv4nodejs/docs/Mat/) 里找到
-
-## 维护者
-
-- [Pillars Zhang](https://github.com/PillarsZhang)
-
-## 感谢
-
-- [opencv4nodejs](https://github.com/justadudewhohacks/opencv4nodejs)
-- [tesseract.js](https://github.com/naptha/tesseract.js)
-
-## License
-
-- [MIT](https://opensource.org/licenses/MIT)
+    - 第三种验证码如果进行训练识别效果会更好
+- sharp
+    - Github | https://github.com/lovell/sharp
+    - API | https://sharp.pixelplumbing.com/api-constructor
+    - 另外吐槽：只有编辑图像的基础功能，而且体验不是很理想需要绕过bug
+- ./lib/fakeOpenCV
+    - 个人仿照 OpenCV 重写了一些图像算法
+
+C++ / Python 的 OpenCV 海量资料也非常有帮助， 相应的函数基本都能在 [opencv4nodejs 的 API 文档](https://justadudewhohacks.github.io/opencv4nodejs/docs/Mat/) 里找到

codes/dots_and_chars/index.js

@@ -0,0 +1,46 @@
+const tesseract_ocr = require("./tesseract_ocr");
+const sharp_cv = require("./sharp_cv");
+
+var ocr;
+const cv = sharp_cv;
+
+class dots_and_chars {
+    recognize = async (image) =>{
+        var timeBegin = Date.now();
+        var cvResult = await cv(image);
+        if (debugFlag) {
+            let cvDebugInfo = {result: "length=" + cvResult.result.length, time : cvResult.time}
+            console.log("cv", cvDebugInfo)
+        }
+
+        var charPromise = [];
+        cvResult.result.forEach((value, index) => {
+            charPromise[index] = ocr.recognize(value); //console.log(value);
+        })
+
+        var charList = await Promise.all(charPromise);
+        charList.forEach((value, index) => {
+            if (['1', 'I'].includes(value.result)){
+                console.log(`index: ${index}, char: ${value.result}, w: ${cvResult.marks[index].w}, h: ${cvResult.marks[index].h}, h/w: ${cvResult.marks[index].h/cvResult.marks[index].w}`);
+                if (cvResult.marks[index].h/cvResult.marks[index].w < 2.2) value.result = '1'
+                  else value.result = 'I';
+            }
+            if (['0', 'O'].includes(value.result)){
+                console.log(`index: ${index}, char: ${value.result}, w: ${cvResult.marks[index].w}, h: ${cvResult.marks[index].h}, h/w: ${cvResult.marks[index].h/cvResult.marks[index].w}`);
+                if (cvResult.marks[index].h/cvResult.marks[index].w < 1.1) value.result = 'O'
+                  else value.result = '0';
+            }
+        })
+
+        if (debugFlag) console.log(charList);
+        var chars = charList.map((value, index) => value.result);
+        return { result: chars.join(''), time: Date.now() - timeBegin };
+    }
+
+    init = async (workers = 4) =>{
+        ocr = new tesseract_ocr(workers)
+        await ocr.init();
+    }
+}
+
+module.exports = new dots_and_chars();

codes/dots_and_chars/sharp_cv.js

@@ -0,0 +1,169 @@
+const cv = require('sharp');
+const fakeOpenCV = require("../../lib/fakeOpenCV");
+const fs = require("fs");
+
+const main = async (file) => {
+    var timeBegin = Date.now();
+    var image = [];
+    image[0] = cv(file);
+    var meta = await image[0].metadata();
+    //console.log(meta);
+    var size = {width: meta.width, height: meta.height};
+
+    //裁边
+    image[1] = image[0].clone().extract({ left: 1, top: 1, width: meta.width-2, height: meta.height-2})
+    size.width -= 2;
+    size.height -= 2;
+    //await showImage(image[1], '裁边');
+
+    //灰度
+    image[2] = image[1].clone().greyscale();
+    //debugFlag && (await showImage(image[2], '灰度'));
+
+    //二值化
+    image[3] = image[2].clone().threshold(255).toColourspace('b-w');
+    debugFlag && (await showImage(image[3], '二值化'));
+
+    //空穴填充
+    image[5] = image[3].clone().linear(-1, 255).convolve({
+        width: 3,
+        height: 3,
+        kernel: [
+            0, 1/4, 0, 
+            1/4, 1, 1/4, 
+            0, 1/4, 0
+        ],
+    })
+    image[5] = await superDeepCopySharp(image[5]);
+    image[5] = image[5].threshold(128).toColourspace('b-w').linear(-1, 255);
+    debugFlag && (await showImage(image[5], '空穴填充'));
+
+    //形态学膨胀
+    var buffer = await image[5].clone().raw().toBuffer();
+    var element = fakeOpenCV.math.ones(3, 3);
+    var imageMath = fakeOpenCV.transformMath({buffer, size})
+    var res = fakeOpenCV.dilate(imageMath, element);
+    var resBuffer = fakeOpenCV.transformBuffer(res);
+    image[4] = await cv(resBuffer.buffer, { raw: { width: resBuffer.size.width, height:resBuffer.size.height, channels: 1 }});
+    debugFlag && (await showImage(image[4], '膨胀'));
+
+    //连通域
+    var res2 = fakeOpenCV.connectedComponents(fakeOpenCV.inverse(res));
+
+    //分割及演示
+    var points = [];
+    var index2 = 0;
+    var marks = [];
+    res2.marks.forEach((value, index) => {
+        var pos = value.position;
+        function setPoints(index2, pos){
+            points[index2] = [
+                [pos.y, pos.x],
+                [pos.y + pos.h - 1, pos.x],
+                [pos.y + pos.h - 1, pos.x + pos.w - 1],
+                [pos.y, pos.x + pos.w - 1],
+                [pos.y, pos.x]
+            ];
+            marks[index2] = pos;
+            marks[index2].area = pos.w * pos.h;
+            marks[index2].cX = Math.floor(pos.x + pos.w / 2);
+            marks[index2].cY = Math.floor(pos.y + pos.h / 2);
+        }
+        //针对超宽的目标块二次对半分割
+        if (pos.w > pos.h * 1.5){
+            pos = {x: pos.x, y: pos.y, w: Math.floor(pos.w / 2), h: pos.h};
+            setPoints(index2++, pos);
+            pos = {x: pos.x + pos.w, y: pos.y, w: pos.w, h: pos.h};
+            setPoints(index2++, pos);
+        } else{
+            setPoints(index2++, pos);
+        }
+    });
+    //points = [[[0, 0], [0, 20], [10, 20], [10, 0], [0, 0]]]
+    var res3 = fakeOpenCV.rectangle(res, points);
+    var resBuffer = fakeOpenCV.transformBuffer(res3);
+    image[5] = cv(resBuffer.buffer, { raw: { width: resBuffer.size.width, height:resBuffer.size.height, channels: 1 }});
+    //debugFlag && (await showImage(image[5], '分割'));
+    //console.log(res2.marks);
+
+    //取出字符块
+    var imageList = [];
+    marks.sort((a, b) => b.area - a.area);
+    marks.splice(4);
+    marks.sort((a, b) => a.cX - b.cX);
+    var w = 45, h = 50;
+    for (let i = 0; i < marks.length; i++){
+        let imageExt = image[4]
+          .clone()
+          .extract({ left: marks[i].x, top: marks[i].y, width: marks[i].w, height: marks[i].h })
+          .extend({
+            top: Math.floor((h - marks[i].h) / 2),
+            bottom: h - Math.floor((h - marks[i].h) / 2) - marks[i].h,
+            left: Math.floor((w - marks[i].w) / 2),
+            right: w - Math.floor((w - marks[i].w) / 2) - marks[i].w,
+            background: "white"
+          });
+        let imageJPEG = await imageExt.jpeg({
+            quality: 100,
+            chromaSubsampling: '4:4:4'
+        }).toBuffer();
+        imageList[i] = imageJPEG;
+        debugFlag && (await showImage(imageExt, i));
+    }
+    return { result: imageList, time: Date.now() - timeBegin, marks };
+}
+
+const showImage = async (image, name) => {
+    var scale = 10;
+    var imageCopy = await superDeepCopySharp(image);
+    var meta = await imageCopy.metadata();
+    imageCopy
+      .clone()
+      .resize({
+        width: meta.width * scale,
+        kernel: cv.kernel.nearest
+      })
+      .jpeg({
+        quality: 100,
+        chromaSubsampling: '4:4:4'
+      })
+      .toBuffer()
+      .then( data => {
+        var tmp = require('tmp');
+        var tmpobj = tmp.dirSync({prefix: 'sharp_' });
+        debugFlag && console.log('Dir: ', tmpobj.name);
+        var join = require("path").join;
+        var exec = require('child_process').exec;
+        var tempPNGPath = join(tmpobj.name, `${name}.jpg`);
+        fs.writeFileSync(tempPNGPath, data);
+        exec(`explorer.exe "${tempPNGPath}"`);
+      })
+      .catch( err => {console.error(err)});
+}
+
+const superDeepCopySharp = async (image) => {
+    var { data, info } = await image.raw().toBuffer({ resolveWithObject: true });
+    var pixelArray = new Uint8ClampedArray(data.buffer);
+    var { width, height, channels } = info;
+    return cv(pixelArray, { raw: { width, height, channels } });
+}
+
+module.exports = main;
+var debugFlag = false;
+
+
+//debugFlag = true;
+if (debugFlag) {
+    //let code = "TTKO";
+    //let code = "54JY";
+    //let code = "7RVO";
+    //let code = "R796";
+    //let code = "XP1R";
+    //let code = "XSVG";
+    //let code = "XQKA";
+    let code = "I8YO";
+    let file = fs.readFileSync(`./examples/${code}.gif`);
+    main(file)
+      .then(res => console.log(res))
+      .catch(err => {console.error(err)});
+};

codes/dots_and_chars/tesseract_ocr.js

@@ -0,0 +1,38 @@
+const { createWorker, createScheduler, PSM, OEM } = require('tesseract.js');
+const scheduler = createScheduler();
+
+class tesseract_ocr {
+    constructor(workers) {
+        this.workersNum = workers
+    };
+    init = async () => {
+        for (var i = 0; i < this.workersNum; i++) {
+            const worker = createWorker();
+            await worker.load();
+            await worker.loadLanguage('eng');
+            await worker.initialize('eng');
+            await worker.setParameters({
+                tessedit_char_whitelist: '0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ',
+                tessedit_pageseg_mode: PSM.SINGLE_CHAR,
+                tessedit_ocr_engine_mode: OEM.TESSERACT_LSTM_COMBINED,
+                //tessedit_ocr_engine_mode: OEM.LSTM_ONLY,
+            });
+            scheduler.addWorker(worker);
+            console.log(`${i + 1}/${this.workersNum} worker(s) initalized`);
+        }
+    };
+    recognize = async (image) => {
+        var timeBegin = Date.now();
+        const { data: { text } } = await scheduler.addJob('recognize', image);
+        return { result: text.replace("\n", ""), time: Date.now() - timeBegin };
+    };
+    terminate = async () => {
+        await scheduler.terminate();
+    };
+    getQueueLen = async () => {
+        var len = await scheduler.getQueueLen();
+        return len;
+    };
+}
+
+module.exports = tesseract_ocr;

judge_and_test.js

@@ -9,14 +9,15 @@ global.debugFlag = 1;
 var modeList = {
     "simplest" : (ans, rightAns) => ans.result == rightAns,
     "grids_and_equations" : (ans, rightAns) => ans.equation.slice(0, 3) == rightAns,
+    "dots_and_chars" : (ans, rightAns) => ans.result == rightAns
 };
 
 (async () => {
     var modeI = 0;
     for(let mode in modeList){
         let cvocr = new cvocrModule(mode);
         console.log(`--- ${++modeI}. ${mode} ---\n`);
-        await cvocr.init(2, 1);
+        await cvocr.init(4, 2);
 
         let examplePath = path.join(__dirname, "codes", mode, "examples");
         let files = fs.readdirSync(examplePath);