淺談OCR之Tesseract
光學字符識別(OCR,Optical Character Recognition)是指對文本資料進行掃描，然后對圖像文件進行分析處理，獲取文字及版面信息的過程。OCR技術非常專業，一般多是印刷、打印行業的從業人員使用，可以快速的將紙質資料轉換為電子資料。關于中文OCR，目前國內水平較高的有清華文通、漢王、尚書，其產品各有千秋，價格不菲。國外OCR發展較早，像一些大公司，如IBM、微軟、HP等，即使沒有推出單獨的OCR產品，但是他們的研發團隊早已掌握核心技術，將OCR功能植入了自身的軟件系統。對于我們程序員來說，一般用不到那么高級的，主要在開發中能夠集成基本的OCR功能就可以了。這兩天我查找了很多免費OCR軟件、類庫，特地整理一下，今天首先來談談Tesseract，下一次將討論下Onenote 2010中的OCR API實現。可以在這里查看OCR技術的發展簡史。
轉載請注明出處：http://www.rzrgm.cn/brooks-dotnet/archive/2010/10/05/1844203.html
1、Tesseract概述
Tesseract的OCR引擎最先由HP實驗室于1985年開始研發，至1995年時已經成為OCR業內最準確的三款識別引擎之一。然而，HP不久便決定放棄OCR業務，Tesseract也從此塵封。
數年以后，HP意識到，與其將Tesseract束之高閣，不如貢獻給開源軟件業，讓其重煥新生－－2005年，Tesseract由美國內華達州信息技術研究所獲得，并求諸于Google對Tesseract進行改進、消除Bug、優化工作。
Tesseract目前已作為開源項目發布在Google Project，其項目主頁在這里查看，其最新版本3.0已經支持中文OCR，并提供了一個命令行工具。本次我們來測試一下Tesseract 3.0，由于命令行對最終用戶不太友好，我用WPF簡單封裝了一下，就可以方便的進行中文OCR了。
1.1、首先到Tesseract項目主頁下載命令行工具、源代碼、中文語言包：
1.2、命令行工具解壓縮后如下（不含1.jpg、1.txt）：
1.3、為了進行中文OCR，將簡體中文語言包復制到【tessdata】目錄下：
1.4、在DOS下切換到Tesseract的命令行目錄，查看一下tesseract.exe的命令格式：
Imagename為待OCR的圖片，outputbase為OCR后的輸出文件，默認是文本文件（.txt），lang為使用的語言包，configfile為配置文件。
1.5、下面來測試一下，準備一張jpg格式的圖片，這里我是放到了和Tesseract同一個目錄中：
輸入：tesseract.exe 1.jpg 1 -l chi_sim，然后回車，幾秒鐘就OCR完成了：
這里注意命令的格式：imagename要加上擴展名.jpg，輸出文件和語言包不需要加擴展名。
OCR結果：
可以看到結果不是很理想，中文識別還說的過去，但是英文、數字大都亂碼。不過作為老牌的OCR引擎，能做到這種程度已經相當不錯了，期待Google的后續升級吧，支持一下。
2、使用WPF封裝Tesseract命令行
2.1、鑒于命令行書寫容易出錯，且對最終用戶很不友好，我做了一個簡單的WPF小程序，將Tesseract的命令行封裝了一下：
左邊選擇圖片、預覽，右邊選擇輸出目錄，顯示OCR結果，支持本地及網絡圖片的預覽。
2.2、為了使得圖片預覽支持縮放、移動，原本打算使用微軟的Zoom It API，可惜不支持WPF，于是使用了一個第三方的類：
圖片縮放、移動工具類
using System;
using System.Windows.Controls;
using System.Windows.Input;
using System.Windows.Media.Animation;
using System.Windows;
using System.Windows.Media;
namespace PanAndZoom
{
public class PanAndZoomViewer : ContentControl
{
public double DefaultZoomFactor { get; set; }
private FrameworkElement source;
private Point ScreenStartPoint = new Point(0, 0);
private TranslateTransform translateTransform;
private ScaleTransform zoomTransform;
private TransformGroup transformGroup;
private Point startOffset;
public PanAndZoomViewer()
{
this.DefaultZoomFactor = 1.4;
}
public override void OnApplyTemplate()
{
base.OnApplyTemplate();
Setup(this);
}
void Setup(FrameworkElement control)
{
this.source = VisualTreeHelper.GetChild(this, 0) as FrameworkElement;
this.translateTransform = new TranslateTransform();
this.zoomTransform = new ScaleTransform();
this.transformGroup = new TransformGroup();
this.transformGroup.Children.Add(this.zoomTransform);
this.transformGroup.Children.Add(this.translateTransform);
this.source.RenderTransform = this.transformGroup;
this.Focusable = true;
this.KeyDown += new KeyEventHandler(source_KeyDown);
this.MouseMove += new MouseEventHandler(control_MouseMove);
this.MouseDown += new MouseButtonEventHandler(source_MouseDown);
this.MouseUp += new MouseButtonEventHandler(source_MouseUp);
this.MouseWheel += new MouseWheelEventHandler(source_MouseWheel);
}
void source_KeyDown(object sender, KeyEventArgs e)
{
// hit escape to reset everything
if (e.Key == Key.Escape) Reset();
}
void source_MouseWheel(object sender, MouseWheelEventArgs e)
{
// zoom into the content. Calculate the zoom factor based on the direction of the mouse wheel.
double zoomFactor = this.DefaultZoomFactor;
if (e.Delta <= 0) zoomFactor = 1.0 / this.DefaultZoomFactor;
// DoZoom requires both the logical and physical location of the mouse pointer
var physicalPoint = e.GetPosition(this);
DoZoom(zoomFactor, this.transformGroup.Inverse.Transform(physicalPoint), physicalPoint);
}
void source_MouseUp(object sender, MouseButtonEventArgs e)
{
if (this.IsMouseCaptured)
{
// we're done. reset the cursor and release the mouse pointer
this.Cursor = Cursors.Arrow;
this.ReleaseMouseCapture();
}
}
void source_MouseDown(object sender, MouseButtonEventArgs e)
{
// Save starting point, used later when determining how much to scroll.
this.ScreenStartPoint = e.GetPosition(this);
this.startOffset = new Point(this.translateTransform.X, this.translateTransform.Y);
this.CaptureMouse();
this.Cursor = Cursors.ScrollAll;
}
void control_MouseMove(object sender, MouseEventArgs e)
{
if (this.IsMouseCaptured)
{
// if the mouse is captured then move the content by changing the translate transform.
// use the Pan Animation to animate to the new location based on the delta between the
// starting point of the mouse and the current point.
var physicalPoint = e.GetPosition(this);
this.translateTransform.BeginAnimation(TranslateTransform.XProperty, CreatePanAnimation(physicalPoint.X - this.ScreenStartPoint.X + this.startOffset.X), HandoffBehavior.Compose);
this.translateTransform.BeginAnimation(TranslateTransform.YProperty, CreatePanAnimation(physicalPoint.Y - this.ScreenStartPoint.Y + this.startOffset.Y), HandoffBehavior.Compose);
}
}
/// <summary>Helper to create the panning animation for x,y coordinates.</summary>
/// <param name="toValue">New value of the coordinate.</param>
/// <returns>Double animation</returns>
private DoubleAnimation CreatePanAnimation(double toValue)
{
var da = new DoubleAnimation(toValue, new Duration(TimeSpan.FromMilliseconds(300)));
da.AccelerationRatio = 0.1;
da.DecelerationRatio = 0.9;
da.FillBehavior = FillBehavior.HoldEnd;
da.Freeze();
return da;
}
/// <summary>Helper to create the zoom double animation for scaling.</summary>
/// <param name="toValue">Value to animate to.</param>
/// <returns>Double animation.</returns>
private DoubleAnimation CreateZoomAnimation(double toValue)
{
var da = new DoubleAnimation(toValue, new Duration(TimeSpan.FromMilliseconds(500)));
da.AccelerationRatio = 0.1;
da.DecelerationRatio = 0.9;
da.FillBehavior = FillBehavior.HoldEnd;
da.Freeze();
return da;
}
/// <summary>Zoom into or out of the content.</summary>
/// <param name="deltaZoom">Factor to mutliply the zoom level by. </param>
/// <param name="mousePosition">Logical mouse position relative to the original content.</param>
/// <param name="physicalPosition">Actual mouse position on the screen (relative to the parent window)</param>
public void DoZoom(double deltaZoom, Point mousePosition, Point physicalPosition)
{
double currentZoom = this.zoomTransform.ScaleX;
currentZoom *= deltaZoom;
this.translateTransform.BeginAnimation(TranslateTransform.XProperty, CreateZoomAnimation(-1 * (mousePosition.X * currentZoom - physicalPosition.X)));
this.translateTransform.BeginAnimation(TranslateTransform.YProperty, CreateZoomAnimation(-1 * (mousePosition.Y * currentZoom - physicalPosition.Y)));
this.zoomTransform.BeginAnimation(ScaleTransform.ScaleXProperty, CreateZoomAnimation(currentZoom));
this.zoomTransform.BeginAnimation(ScaleTransform.ScaleYProperty, CreateZoomAnimation(currentZoom));
}
/// <summary>Reset to default zoom level and centered content.</summary>
public void Reset()
{
this.translateTransform.BeginAnimation(TranslateTransform.XProperty, CreateZoomAnimation(0));
this.translateTransform.BeginAnimation(TranslateTransform.YProperty, CreateZoomAnimation(0));
this.zoomTransform.BeginAnimation(ScaleTransform.ScaleXProperty, CreateZoomAnimation(1));
this.zoomTransform.BeginAnimation(ScaleTransform.ScaleYProperty, CreateZoomAnimation(1));
}
}
}
using System.Windows.Controls;
using System.Windows.Input;
using System.Windows.Media.Animation;
using System.Windows;
using System.Windows.Media;
namespace PanAndZoom
{
public class PanAndZoomViewer : ContentControl
{
public double DefaultZoomFactor { get; set; }
private FrameworkElement source;
private Point ScreenStartPoint = new Point(0, 0);
private TranslateTransform translateTransform;
private ScaleTransform zoomTransform;
private TransformGroup transformGroup;
private Point startOffset;
public PanAndZoomViewer()
{
this.DefaultZoomFactor = 1.4;
}
public override void OnApplyTemplate()
{
base.OnApplyTemplate();
Setup(this);
}
void Setup(FrameworkElement control)
{
this.source = VisualTreeHelper.GetChild(this, 0) as FrameworkElement;
this.translateTransform = new TranslateTransform();
this.zoomTransform = new ScaleTransform();
this.transformGroup = new TransformGroup();
this.transformGroup.Children.Add(this.zoomTransform);
this.transformGroup.Children.Add(this.translateTransform);
this.source.RenderTransform = this.transformGroup;
this.Focusable = true;
this.KeyDown += new KeyEventHandler(source_KeyDown);
this.MouseMove += new MouseEventHandler(control_MouseMove);
this.MouseDown += new MouseButtonEventHandler(source_MouseDown);
this.MouseUp += new MouseButtonEventHandler(source_MouseUp);
this.MouseWheel += new MouseWheelEventHandler(source_MouseWheel);
}
void source_KeyDown(object sender, KeyEventArgs e)
{
// hit escape to reset everything
if (e.Key == Key.Escape) Reset();
}
void source_MouseWheel(object sender, MouseWheelEventArgs e)
{
// zoom into the content. Calculate the zoom factor based on the direction of the mouse wheel.
double zoomFactor = this.DefaultZoomFactor;
if (e.Delta <= 0) zoomFactor = 1.0 / this.DefaultZoomFactor;
// DoZoom requires both the logical and physical location of the mouse pointer
var physicalPoint = e.GetPosition(this);
DoZoom(zoomFactor, this.transformGroup.Inverse.Transform(physicalPoint), physicalPoint);
}
void source_MouseUp(object sender, MouseButtonEventArgs e)
{
if (this.IsMouseCaptured)
{
// we're done. reset the cursor and release the mouse pointer
this.Cursor = Cursors.Arrow;
this.ReleaseMouseCapture();
}
}
void source_MouseDown(object sender, MouseButtonEventArgs e)
{
// Save starting point, used later when determining how much to scroll.
this.ScreenStartPoint = e.GetPosition(this);
this.startOffset = new Point(this.translateTransform.X, this.translateTransform.Y);
this.CaptureMouse();
this.Cursor = Cursors.ScrollAll;
}
void control_MouseMove(object sender, MouseEventArgs e)
{
if (this.IsMouseCaptured)
{
// if the mouse is captured then move the content by changing the translate transform.
// use the Pan Animation to animate to the new location based on the delta between the
// starting point of the mouse and the current point.
var physicalPoint = e.GetPosition(this);
this.translateTransform.BeginAnimation(TranslateTransform.XProperty, CreatePanAnimation(physicalPoint.X - this.ScreenStartPoint.X + this.startOffset.X), HandoffBehavior.Compose);
this.translateTransform.BeginAnimation(TranslateTransform.YProperty, CreatePanAnimation(physicalPoint.Y - this.ScreenStartPoint.Y + this.startOffset.Y), HandoffBehavior.Compose);
}
}
/// <summary>Helper to create the panning animation for x,y coordinates.</summary>
/// <param name="toValue">New value of the coordinate.</param>
/// <returns>Double animation</returns>
private DoubleAnimation CreatePanAnimation(double toValue)
{
var da = new DoubleAnimation(toValue, new Duration(TimeSpan.FromMilliseconds(300)));
da.AccelerationRatio = 0.1;
da.DecelerationRatio = 0.9;
da.FillBehavior = FillBehavior.HoldEnd;
da.Freeze();
return da;
}
/// <summary>Helper to create the zoom double animation for scaling.</summary>
/// <param name="toValue">Value to animate to.</param>
/// <returns>Double animation.</returns>
private DoubleAnimation CreateZoomAnimation(double toValue)
{
var da = new DoubleAnimation(toValue, new Duration(TimeSpan.FromMilliseconds(500)));
da.AccelerationRatio = 0.1;
da.DecelerationRatio = 0.9;
da.FillBehavior = FillBehavior.HoldEnd;
da.Freeze();
return da;
}
/// <summary>Zoom into or out of the content.</summary>
/// <param name="deltaZoom">Factor to mutliply the zoom level by. </param>
/// <param name="mousePosition">Logical mouse position relative to the original content.</param>
/// <param name="physicalPosition">Actual mouse position on the screen (relative to the parent window)</param>
public void DoZoom(double deltaZoom, Point mousePosition, Point physicalPosition)
{
double currentZoom = this.zoomTransform.ScaleX;
currentZoom *= deltaZoom;
this.translateTransform.BeginAnimation(TranslateTransform.XProperty, CreateZoomAnimation(-1 * (mousePosition.X * currentZoom - physicalPosition.X)));
this.translateTransform.BeginAnimation(TranslateTransform.YProperty, CreateZoomAnimation(-1 * (mousePosition.Y * currentZoom - physicalPosition.Y)));
this.zoomTransform.BeginAnimation(ScaleTransform.ScaleXProperty, CreateZoomAnimation(currentZoom));
this.zoomTransform.BeginAnimation(ScaleTransform.ScaleYProperty, CreateZoomAnimation(currentZoom));
}
/// <summary>Reset to default zoom level and centered content.</summary>
public void Reset()
{
this.translateTransform.BeginAnimation(TranslateTransform.XProperty, CreateZoomAnimation(0));
this.translateTransform.BeginAnimation(TranslateTransform.YProperty, CreateZoomAnimation(0));
this.zoomTransform.BeginAnimation(ScaleTransform.ScaleXProperty, CreateZoomAnimation(1));
this.zoomTransform.BeginAnimation(ScaleTransform.ScaleYProperty, CreateZoomAnimation(1));
}
}
}
2.3、除了使用鼠標。還可以使用滾動條調節圖片預覽效果：
數據綁定 <WrapPanel Grid.Row="2" Grid.Column="0">
<Label Name="lab長度" Content="長度：" Margin="3" />
<Slider Name="sl長度" MinWidth="50" Margin="3" VerticalAlignment="Center" Maximum="400" Value="{Binding ElementName=img圖片, Path=Width, Mode=TwoWay}" />
<Label Name="lab寬度" Content="寬度：" Margin="3" />
<Slider Name="sl寬度" MinWidth="50" Margin="3" VerticalAlignment="Center" Maximum="400" Value="{Binding ElementName=img圖片, Path=Height, Mode=TwoWay}" />
<Label Name="lab透明度" Content="透明度：" Margin="3" />
<Slider Name="sl透明度" MinWidth="50" Margin="3" VerticalAlignment="Center" Maximum="1" Value="{Binding ElementName=img圖片, Path=Opacity, Mode=TwoWay}" />
<Label Name="lab拉伸方式" Content="拉伸方式：" Margin="3" />
<ComboBox Name="txt拉伸方式" Margin="3" MinWidth="85">
<ComboBoxItem Content="Fill" />
<ComboBoxItem Content="None" IsSelected="True" />
<ComboBoxItem Content="Uniform" />
<ComboBoxItem Content="UniformToFill" />
</ComboBox>
</WrapPanel>
<local:PanAndZoomViewer Grid.Row="3" Grid.Column="0" Height="300" Margin="3">
<Image Name="img圖片" Stretch="{Binding ElementName=txt拉伸方式, Path=Text, Mode=TwoWay}" />
</local:PanAndZoomViewer>
<Label Name="lab長度" Content="長度：" Margin="3" />
<Slider Name="sl長度" MinWidth="50" Margin="3" VerticalAlignment="Center" Maximum="400" Value="{Binding ElementName=img圖片, Path=Width, Mode=TwoWay}" />
<Label Name="lab寬度" Content="寬度：" Margin="3" />
<Slider Name="sl寬度" MinWidth="50" Margin="3" VerticalAlignment="Center" Maximum="400" Value="{Binding ElementName=img圖片, Path=Height, Mode=TwoWay}" />
<Label Name="lab透明度" Content="透明度：" Margin="3" />
<Slider Name="sl透明度" MinWidth="50" Margin="3" VerticalAlignment="Center" Maximum="1" Value="{Binding ElementName=img圖片, Path=Opacity, Mode=TwoWay}" />
<Label Name="lab拉伸方式" Content="拉伸方式：" Margin="3" />
<ComboBox Name="txt拉伸方式" Margin="3" MinWidth="85">
<ComboBoxItem Content="Fill" />
<ComboBoxItem Content="None" IsSelected="True" />
<ComboBoxItem Content="Uniform" />
<ComboBoxItem Content="UniformToFill" />
</ComboBox>
</WrapPanel>
<local:PanAndZoomViewer Grid.Row="3" Grid.Column="0" Height="300" Margin="3">
<Image Name="img圖片" Stretch="{Binding ElementName=txt拉伸方式, Path=Text, Mode=TwoWay}" />
</local:PanAndZoomViewer>
2.4、由于Tesseract命令行不支持直接OCR網絡圖片，故先下載：
圖片下載 private void fnStartDownload(string v_strImgPath, string v_strOutputDir, out string v_strTmpPath)
{
int n = v_strImgPath.LastIndexOf('/');
string URLAddress = v_strImgPath.Substring(0, n);
string fileName = v_strImgPath.Substring(n + 1, v_strImgPath.Length - n - 1);
this.__OutputFileName = v_strOutputDir + "\\" + fileName.Substring(0, fileName.LastIndexOf("."));
if (!Directory.Exists(System.Configuration.ConfigurationManager.AppSettings["tmpPath"]))
{
Directory.CreateDirectory(System.Configuration.ConfigurationManager.AppSettings["tmpPath"]);
}
string Dir = System.Configuration.ConfigurationManager.AppSettings["tmpPath"];
v_strTmpPath = Dir + "\\" + fileName;
WebRequest myre = WebRequest.Create(URLAddress);
client.DownloadFile(v_strImgPath, v_strTmpPath);
//Stream str = client.OpenRead(v_strImgPath);
//StreamReader reader = new StreamReader(str);
//byte[] mbyte = new byte[Int32.Parse(System.Configuration.ConfigurationManager.AppSettings["MaxDownloadImgLength"])];
//int allmybyte = (int)mbyte.Length;
//int startmbyte = 0;
//while (allmybyte > 0)
//{
// int m = str.Read(mbyte, startmbyte, allmybyte);
// if (m == 0)
// {
// break;
// }
// startmbyte += m;
// allmybyte -= m;
//}
//FileStream fstr = new FileStream(v_strTmpPath, FileMode.Create, FileAccess.Write);
//fstr.Write(mbyte, 0, startmbyte);
//str.Close();
//fstr.Close();
}
{
int n = v_strImgPath.LastIndexOf('/');
string URLAddress = v_strImgPath.Substring(0, n);
string fileName = v_strImgPath.Substring(n + 1, v_strImgPath.Length - n - 1);
this.__OutputFileName = v_strOutputDir + "\\" + fileName.Substring(0, fileName.LastIndexOf("."));
if (!Directory.Exists(System.Configuration.ConfigurationManager.AppSettings["tmpPath"]))
{
Directory.CreateDirectory(System.Configuration.ConfigurationManager.AppSettings["tmpPath"]);
}
string Dir = System.Configuration.ConfigurationManager.AppSettings["tmpPath"];
v_strTmpPath = Dir + "\\" + fileName;
WebRequest myre = WebRequest.Create(URLAddress);
client.DownloadFile(v_strImgPath, v_strTmpPath);
//Stream str = client.OpenRead(v_strImgPath);
//StreamReader reader = new StreamReader(str);
//byte[] mbyte = new byte[Int32.Parse(System.Configuration.ConfigurationManager.AppSettings["MaxDownloadImgLength"])];
//int allmybyte = (int)mbyte.Length;
//int startmbyte = 0;
//while (allmybyte > 0)
//{
// int m = str.Read(mbyte, startmbyte, allmybyte);
// if (m == 0)
// {
// break;
// }
// startmbyte += m;
// allmybyte -= m;
//}
//FileStream fstr = new FileStream(v_strTmpPath, FileMode.Create, FileAccess.Write);
//fstr.Write(mbyte, 0, startmbyte);
//str.Close();
//fstr.Close();
}
2.5、使用Process來調用Tesseract命令行：
調用Tesseract命令行 private void fnOCR(string v_strTesseractPath, string v_strSourceImgPath, string v_strOutputPath, string v_strLangPath)
{
using (Process process = new System.Diagnostics.Process())
{
process.StartInfo.FileName = v_strTesseractPath;
process.StartInfo.Arguments = v_strSourceImgPath + " " + v_strOutputPath + " -l " + v_strLangPath;
process.StartInfo.UseShellExecute = false;
process.StartInfo.CreateNoWindow = true;
process.StartInfo.RedirectStandardOutput = true;
process.Start();
process.WaitForExit();
}
}
{
using (Process process = new System.Diagnostics.Process())
{
process.StartInfo.FileName = v_strTesseractPath;
process.StartInfo.Arguments = v_strSourceImgPath + " " + v_strOutputPath + " -l " + v_strLangPath;
process.StartInfo.UseShellExecute = false;
process.StartInfo.CreateNoWindow = true;
process.StartInfo.RedirectStandardOutput = true;
process.Start();
process.WaitForExit();
}
}
2.6、測試本地圖片：
2.7、測試網絡圖片：
小結：
本次我們簡單討論了下Tesseract的用法，作為一款開源、免費的OCR引擎，能夠支持中文十分難得。雖然其識別效果不是很理想，但是對于要求不高的中小型項目來說，已經足夠用了。這里有一份免費OCR工具列表，感興趣的朋友可以研究一下。下一次將測試一下Onenote 2010中OCR功能，以及如何調用其API，為項目所用。
目前主要研發檢測機構解決方案 質檢、計量、疾控、環境等 商務請聯系 15952187581(微信同號)
浙公網安備 33010602011771號