[Tesseract] 简单数字识别图像识别涉及的理论:傅里叶变换,图形形态学,滤波,矩阵变换等等.
Tesseract的出现为了解决在没有这些复杂的理论基础,快速识别图像的框架.
准备:
1.样本图像学习,预处理(平均每1个元素出现20次)
2.学习,初步识别
3.校正学习库
测试:
1.待识别图像,预处理
2.根据学习库 识别
例子1:图片反色
1 PRivate static void Reverse(string fileName,string outName) 2 { 3 using (var pic = Image.FromFile(fileName) as Bitmap) 4 { 5 for (int i = 0; i < pic.Width; i++) 6 { 7 for (int j = 0; j < pic.Height; j++) 8 { 9 var c = pic.GetPixel(i, j);10 c = Color.FromArgb(255 - c.R, 255 - c.G, 255 - c.B);11 pic.SetPixel(i, j, c);12 }13 }14 pic.Save(outName);15 }16 }
Reverse
例子2:取一个图片的指定区域
1 private Image Analyse(string fileName)//为方便,方法参数没有用Image 2 { 3 using (var map = Image.FromFile(fileName) as Bitmap) 4 { 5 if (map == null) return null; 6 Point p1; 7 Point p2; 8 var p = GetConfig(out p1, out p2); 9 var pic = new Bitmap(p.X, p.Y);10 var x = 0;11 var y = 0;12 for (int i = 0; i < map.Height; i++)13 {14 if (i >= p1.Y && i <= p2.Y)15 {16 for (int j = 0; j < map.Width; j++)17 {18 if (j >= p1.X && j <= p2.X)19 {20 pic.SetPixel(x, y, map.GetPixel(j, i));21 x++;22 }23 }24 x = 0;25 y++;26 }27 }28 return pic;29 }30 }31 32 private Point GetConfig(out Point p1, out Point p2)33 {34 var p1Str = ConfigurationManager.AppSettings["p1"].Split(',');35 var p2Str = ConfigurationManager.AppSettings["p2"].Split(',');36 p1 = new Point() { X = int.Parse(p1Str[0]), Y = int.Parse(p1Str[1]) };37 p2 = new Point() { X = int.Parse(p2Str[0]), Y = int.Parse(p2Str[1]) };38 return new Point() { X = p2.X - p1.X + 2, Y = p2.Y - p1.Y + 2 };39 }40 41 42 class Point43 {44 /// <summary>45 /// 点的X坐标,或者宽度46 /// </summary>47 public int X { get; set; }48 /// <summary>49 /// 点的Y坐标,或者高度50 /// </summary>51 public int Y { get; set; }52 }
Image Analyse(string fileName)
识别步骤:
1.将图片转成tif格式,通过jTessBoxEditor程序把所有tif图片打包成1个tif图片.
2.安装tesseract-ocr-setup-3.01-1.exe(用安装包版不需要配环境变量).
3.dos命令:输入tesseract.exe {0}.tif {0} batch.nochop makebox生成box文件({0}为文件名)
4.dos命令:输入tesseract.exe {0}.tif {0}nobatch box.train生成tr文件
5.dos命令:输入unicharset_extractor.exe tj.box生成unicharset文件
6.同一目录下新建一个文件font_properties无扩展名.写入{0}1 0 0 1 0
7.dos命令:输入cntraining.exe {0}.tr
8.dos命令:输入mftraining.exe -F font_properties -U unicharset {0}.tr
9.4个文件名加前缀:{0}. (1.有.2.4个:unicharset inttemp normproto pfftable)
10:dos命令:combine_tessdata {0}.(合并所有文件,生成学习库,{0}.traineddata)
代码步骤:
1.解压tesseractdotnet_v301_r590.zip,添加tesseract.dll引用
2.
var tp= new TesseractProcessor();
tp.SetPageSegMode(ePageSegMode.PSM_SINGLE_LINE);
tp.Init("{0}","{1}", (int)eOcrEngineMode.OEM_DEFAULT);//{0}为Path名,如c:\test\,{1}为学习库文件名,不需要扩展名.
var result = tp.Recognize(pic);
3.生成目标平台x86.
工具下载:点击下载
