JAVA识别图形验证码

最近一直在开发一个用于自动发帖的工具，用HttpClient模拟客户端浏览器注册发帖。但是碰到了图形验证码的问题了，对单数字的验证码，通过一些OCR引擎，如：tesseract，AspriseOCR很容易解决问题。但碰到如CSDN论坛这中图形验证码就比较麻烦，必须先通过预处理。使图象二值化，黑白灰度，增加亮度。我的代码如下：package myfilter; import java.io.*; import java.awt.image.*; import java.awt.geom.AffineTransform; import java.awt.color.ColorSpace; import java.awt.image.ConvolveOp; import java.awt.image.Kernel; import java.awt.image.BufferedImage; import javax.imageio.ImageIO; import java.awt.Toolkit; import java.awt.Image; /** * <p>Title: Image Filter</p> * * <p>Description: image processing by filters </p> * <p>Copyright: Copyright (c) 2010</p> * * @author gl74gs48@163.com * @since jdk1.5.0 * @version 1.0 */ public class MyImgFilter { BufferedImage image; private int iw, ih; private int[] pixels; public MyImgFilter(BufferedImage image) { this.image = image; iw = image.getWidth(); ih = image.getHeight(); pixels = new int[iw * ih]; } /** 图像二值化 */ public BufferedImage changeGrey() { PixelGrabber pg = new PixelGrabber(image.getSource(), 0, 0, iw, ih, pixels,0, iw); try { pg.grabPixels(); } catch (InterruptedException e) { e.printStackTrace(); } // 设定二值化的域值，默认值为100 int grey = 100; // 对图像进行二值化处理，Alpha值保持不变 ColorModel cm = ColorModel.getRGBdefault(); for (int i = 0; i < iw * ih; i++) { int red, green, blue; int alpha = cm.getAlpha(pixels[i]); if (cm.getRed(pixels[i]) > grey) { red = 255; } else { red = 0; } if (cm.getGreen(pixels[i]) > grey) { green = 255; } else { green = 0; } if (cm.getBlue(pixels[i]) > grey) { blue = 255; } else { blue = 0; } pixels[i] = alpha << 24 | red << 16 | green << 8 | blue; //通过移位重新构成某一点像素的RGB值 } // 将数组中的象素产生一个图像 Image tempImg=Toolkit.getDefaultToolkit().createImage(new MemoryImageSource(iw,ih, pixels, 0, iw)); image = new BufferedImage(tempImg.getWidth(null),tempImg.getHeight(null), BufferedImage.TYPE_INT_BGR ); image.createGraphics().drawImage(tempImg, 0, 0, null); return image; } /** 中值滤波 */ public BufferedImage getMedian() { PixelGrabber pg = new PixelGrabber(image.getSource(), 0, 0, iw, ih, pixels, 0, iw); try { pg.grabPixels(); } catch (InterruptedException e) { e.printStackTrace(); } // 对图像进行中值滤波，Alpha值保持不变 ColorModel cm = ColorModel.getRGBdefault(); for (int i = 1; i < ih - 1; i++) { for (int j = 1; j < iw - 1; j++) { int red, green, blue; int alpha = cm.getAlpha(pixels[i * iw + j]); // int red2 = cm.getRed(pixels[(i - 1) * iw + j]); int red4 = cm.getRed(pixels[i * iw + j - 1]); int red5 = cm.getRed(pixels[i * iw + j]); int red6 = cm.getRed(pixels[i * iw + j + 1]); // int red8 = cm.getRed(pixels[(i + 1) * iw + j]); // 水平方向进行中值滤波 if (red4 >= red5) { if (red5 >= red6) { red = red5; } else { if (red4 >= red6) { red = red6; } else { red = red4; } } } else { if (red4 > red6) { red = red4; } else { if (red5 > red6) { red = red6; } else { red = red5; } } } int green4 = cm.getGreen(pixels[i * iw + j - 1]); int green5 = cm.getGreen(pixels[i * iw + j]); int green6 = cm.getGreen(pixels[i * iw + j + 1]); // 水平方向进行中值滤波 if (green4 >= green5) { if (green5 >= green6) { green = green5; } else { if (green4 >= green6) { green = green6; } else { green = green4; } } } else { if (green4 > green6) { green = green4; } else { if (green5 > green6) { green = green6; } else { green = green5; } } } // int blue2 = cm.getBlue(pixels[(i - 1) * iw + j]); int blue4 = cm.getBlue(pixels[i * iw + j - 1]); int blue5 = cm.getBlue(pixels[i * iw + j]); int blue6 = cm.getBlue(pixels[i * iw + j + 1]); // int blue8 = cm.getBlue(pixels[(i + 1) * iw + j]); // 水平方向进行中值滤波 if (blue4 >= blue5) { if (blue5 >= blue6) { blue = blue5; } else { if (blue4 >= blue6) { blue = blue6; } else { blue = blue4; } } } else { if (blue4 > blue6) { blue = blue4; } else { if (blue5 > blue6) { blue = blue6; } else { blue = blue5; } } } pixels[i * iw + j] = alpha << 24 | red << 16 | green << 8 | blue; } } // 将数组中的象素产生一个图像 Image tempImg=Toolkit.getDefaultToolkit().createImage(new MemoryImageSource(iw,ih, pixels, 0, iw)); image = new BufferedImage(tempImg.getWidth(null),tempImg.getHeight(null), BufferedImage.TYPE_INT_BGR ); image.createGraphics().drawImage(tempImg, 0, 0, null); return image; } public BufferedImage getGrey() { ColorConvertOp ccp=new ColorConvertOp(ColorSpace.getInstance(ColorSpace.CS_GRAY), null); return image=ccp.filter(image,null); } //Brighten using a linear formula that increases all color values public BufferedImage getBrighten() { RescaleOp rop=new RescaleOp(1.25f, 0, null); return image=rop.filter(image,null); } //Blur by "convolving" the image with a matrix public BufferedImage getBlur() { float[] data = { .1111f, .1111f, .1111f, .1111f, .1111f, .1111f, .1111f, .1111f, .1111f, }; ConvolveOp cop = new ConvolveOp(new Kernel(3, 3, data)); return image=cop.filter(image,null); } // Sharpen by using a different matrix public BufferedImage getSharpen() { float[] data = { 0.0f, -0.75f, 0.0f, -0.75f, 4.0f, -0.75f, 0.0f, -0.75f, 0.0f}; ConvolveOp cop = new ConvolveOp(new Kernel(3, 3, data)); return image=cop.filter(image,null); } // 11) Rotate the image 180 degrees about its center point public BufferedImage getRotate() { AffineTransformOp atop=new AffineTransformOp(AffineTransform.getRotateInstance(Math.PI,image.getWidth()/2,image.getHeight()/2), AffineTransformOp.TYPE_NEAREST_NEIGHBOR); return image=atop.filter(image,null); } public BufferedImage getProcessedImg() { return image; } public static void main(String[] args) throws IOException { FileInputStream fin=new FileInputStream(args[0]); BufferedImage bi=ImageIO.read(fin); MyImgFilter flt=new MyImgFilter(bi); flt.changeGrey(); flt.getGrey(); flt.getBrighten(); bi=flt.getProcessedImg(); String pname=args[0].substring(0,args[0].lastIndexOf(".")); File file = new File(pname+".jpg"); ImageIO.write(bi, "jpg", file); } }

运行java myfilter.MyImgFilter t6.bmp，请确认图片t6.bmp与myfilter目录在同一目录下。

顺便说一下，在JDK1.5下，ImageIO可以输出JPG，BMP，PNG三种格式图片，但不支持GIF图片输出。

经处理后图片的识别率大大提高。

部分代码参考http://ykf.javaeye.com/blog/212431及《Java Examples In A Nutshell 3rd》