SkySeraph Jun 8th 2011 HQU
Email:zgzhaobo@gmail.com QQ:452728574
Latest Modified Date:Jun 8th 2011 HQU
一 原理及说明:
1 RGB(red,green,blue)模式是一种与设备相关的色彩空间,最常用的用途就是显示器系统。RGB下,各分量关联性太大,每个通道都编入了亮度信息,容易受周围环境影响(光照等),其与人眼认知颜色的过程不太匹配,并不适合用来对彩色图像进行分析和分割,相比下HSV空间是从人的视觉系统除法的,更适于图像分析等。更多关于各种彩色空间模型请参考http://www.cnblogs.com/skyseraph/archive/2011/05/03/2035643.html
2 国内很多关于车牌识别的论文中,当利用到颜色信息时,一般都是在HSV/YIQ/Lab模式下,根据特定的车牌颜色信息(常见车牌颜色有:白底黑字、黑底白字、蓝底白字、黄底黑字等),进行车牌分割进行的。 颜色的提取方法即本文所述。 这种方法只适合特定颜色的提取,用PR术语,类似"有监督学习";反之,无监督,对任意图像进行颜色分割,属于彩色分割领域。
3 关于HSV范围的划分:
<1> 论文:Car color recognition from CCTV camera image:http://www.docin.com/p-211572110.html
作者采用的是如下方式:
<2>论文:利用支持向量机识别汽车颜色:http://www.cnki.com.cn/Article/CJFDTotal-JSJF200405018.htm
作者首先是在Lab空间下分出16类颜色,然后再HSV下进行样本空间分解,采用如下方式:
<3>本文根据实验,采取划分方式如源码所示,在这种方式下,测试结果较好。
二 源码:
/////////////////////////////////////////////////////////////////////////////
// Note: 颜色分割:提取特定颜色
// Version: 5/11/2011 skyseraph zgzhaobo@gmail.com
/////////////////////////////////////////////////////////////////////////////
void CColorSegDlg::ColorSegByHSV(IplImage* img)
// 提取特定颜色
{
//====================== 变量定义====================//
int x,y; //循环
//====================== 输入彩色图像信息====================//
IplImage* pSrc = NULL;
pSrc = cvCreateImage(cvGetSize(img),img->depth,img->nChannels);
cvCopyImage(img,pSrc);
int width = pSrc->width; //图像宽度
int height = pSrc->height; //图像高度
int depth = pSrc->depth; //图像位深(IPL_DEPTH_8U...)
int channels = pSrc->nChannels; //图像通道数(1、2、3、4)
int imgSize = pSrc->imageSize; //图像大小 imageSize = height*widthStep
int step = pSrc->widthStep/sizeof(uchar); //相邻行的同列点之间的字节数: 注意widthStep != width*nChannels (有字节填零补充)
uchar* data = (uchar *)pSrc->imageData;
int imageLen = width*height; //
//=========================================//
double B=0.0,G=0.0,R=0.0,H=0.0,S=0.0,V=0.0;
IplImage* dstColorSegByColor = cvCreateImage(cvGetSize(pSrc),IPL_DEPTH_8U,3);
IplImage* dstColorSegByColorGray = cvCreateImage(cvGetSize(pSrc),IPL_DEPTH_8U,1);
//CvFont font = cvFont( 1, 1 );
for (y=0; y<height; y++)
{
for ( x=0; x<width; x++)
{
// 获取BGR值
B = ((uchar*)(pSrc->imageData + y*pSrc->widthStep))[x*pSrc->nChannels];
G = ((uchar*)(pSrc->imageData + y*pSrc->widthStep))[x*pSrc->nChannels+1];
R = ((uchar*)(pSrc->imageData + y*pSrc->widthStep))[x*pSrc->nChannels+2];
// RGB-HSV
pMyColorSpace.RGB2HSV(R,G,B,H,S,V);
H = (360*H)/(2*PI);
// 黑白
//黑色
if(V<0.35)
{
((uchar*)(dstColorSegByColorGray->imageData + y*dstColorSegByColorGray->widthStep))[x]
= 0; //灰度
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels]
= 0; //B
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+1]
= 0; //G
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+2]
= 0; //R
}
//白色
if(S<0.15 && V>0.75)
{
((uchar*)(dstColorSegByColorGray->imageData + y*dstColorSegByColorGray->widthStep))[x]
= 255; //灰度
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels]
= 255; //B
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+1]
= 255; //G
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+2]
= 255; //R
}
//灰色
if(S<0.15 && 0.35<V && V<0.75)
{
((uchar*)(dstColorSegByColorGray->imageData + y*dstColorSegByColorGray->widthStep))[x]
= 128; //灰度
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels]
= 128; //B
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+1]
= 128; //G
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+2]
= 128; //R
}
// 彩色
if(V>=0.35 && S>=0.15)
{
//红色相近
if((H>=0 && H<15) || (H>=340 && H<360))
{
((uchar*)(dstColorSegByColorGray->imageData + y*dstColorSegByColorGray->widthStep))[x]
= 40; //灰度
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels]
= 0; //B
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+1]
= 0; //G
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+2]
= 255; //R
}
//黄色相近
else if(H>=15 && H<75)
{
((uchar*)(dstColorSegByColorGray->imageData + y*dstColorSegByColorGray->widthStep))[x]
= 80; //灰度
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels]
= 0; //B
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+1]
= 255; //G
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+2]
= 255; //R
}
//绿色相近
else if(H>=75 && H<150)
{
((uchar*)(dstColorSegByColorGray->imageData + y*dstColorSegByColorGray->widthStep))[x]
= 120; //灰度
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels]
= 0; //B
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+1]
= 255; //G
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+2]
= 0; //R
}
///*//青色相近
else if(H>=150 && H<185)
{
((uchar*)(dstColorSegByColorGray->imageData + y*dstColorSegByColorGray->widthStep))[x]
= 160; //灰度
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels]
= 255; //B
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+1]
= 255; //G
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+2]
= 0; //R
}//*/
//蓝色相近
else if(H>=185 && H<270)
{
((uchar*)(dstColorSegByColorGray->imageData + y*dstColorSegByColorGray->widthStep))[x]
= 200; //灰度
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels]
= 255; //B
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+1]
= 0; //G
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+2]
= 0; //R
}
// /* //洋红:270-340
else if(H>=270 && H<340)
{
((uchar*)(dstColorSegByColorGray->imageData + y*dstColorSegByColorGray->widthStep))[x]
= 220; //灰度
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels]
= 255; //B
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+1]
= 0; //G
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+2]
= 255; //R
}//*/
else
{
((uchar*)(dstColorSegByColorGray->imageData + y*dstColorSegByColorGray->widthStep))[x]
= 180; //灰度
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels]
= 128; //B //紫色Purple
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+1]
= 0; //G
((uchar*)(dstColorSegByColor->imageData + y*dstColorSegByColor->widthStep))[x*dstColorSegByColor->nChannels+2]
= 128; //R
}
}
}
}
//cvNamedWindow("src",1);
//cvShowImage("src",pSrc);
cvNamedWindow("dstColorSegByColor",1);
cvShowImage("dstColorSegByColor",dstColorSegByColor);
cvNamedWindow("dstColorSegByColorGray",1);
cvShowImage("dstColorSegByColorGray",dstColorSegByColorGray);
cvSaveImage(".\\dstColorSegByColor.jpg",dstColorSegByColor);
cvSaveImage(".\\dstColorSegByColorGray.jpg",dstColorSegByColorGray);
cvWaitKey(0);
cvDestroyAllWindows();
cvReleaseImage(&pSrc);
cvReleaseImage(&dstColorSegByColor);
cvReleaseImage(&dstColorSegByColorGray);
}
三 效果:
(1)原图
(2)颜色分割后彩色图
(3)颜色分割后灰度图(利用不同灰度级显示)
四 补充(RGB模式下,来源网络)
1 源码
void CFindRGBDlg::OnFind()
{
int color=m_colorList.GetCurSel();
pic=cvCreateImage( cvSize(image->width,image->height), 8, 1 );
cvZero(pic);
for(int x=0;x<image->height;x++)
{
for(int y=0;y<image->width;y++)
{
uchar* ptrImg = &CV_IMAGE_ELEM(image,uchar,x,y*3);
// uchar* ptrPic = &((uchar*)(pic->imageData + pic->widthStep*y))[x];
//red
if(color==0)
{
if((ptrImg[0]-ptrImg[1])>200&&(ptrImg[0]-ptrImg[2])>200)
CV_IMAGE_ELEM(pic,uchar,x,y)=255;
}
//Green
else if(color==1)
{
if((ptrImg[1]-ptrImg[0])>200&&(ptrImg[1]-ptrImg[2])>200)
CV_IMAGE_ELEM(pic,uchar,x,y)=255;
}
//blue
else if(color==2)
{
if((ptrImg[2]-ptrImg[0])>200&&(ptrImg[2]-ptrImg[1])>200)
CV_IMAGE_ELEM(pic,uchar,x,y)=255;
}
}
}
cvNamedWindow("temp",-1);
cvShowImage("temp",pic);
cvWaitKey();
storage = cvCreateMemStorage(0);
contour = 0;
mode = CV_RETR_EXTERNAL;
cvFindContours( pic, storage, &contour, sizeof(CvContour),
mode, CV_CHAIN_APPROX_SIMPLE);
cvDrawContours(image, contour,
CV_RGB(0,0,0), CV_RGB(0, 0, 0),
2, 2, 8);
CRect rect;
GetDlgItem(IDC_PICTURE)->GetClientRect(&rect);
InvalidateRect(rect,true);
}
2 效果:
More in http://skyseraph.com/2011/08/27/CV/图像算法专题/
Author: SKySeraph
Email/GTalk: zgzhaobo@gmail.com QQ:452728574
From: http://www.cnblogs.com/skyseraph/
本文版权归作者和博客园共有,欢迎转载,但未经作者同意必须保留此段声明,且在文章页面明显位置给出原文连接,请尊重作者的劳动成果