图像特征检测(Image Feature Detection)

作者：王先荣

前言

图像特征提取是计算机视觉和图像处理中的一个概念。它指的是使用计算机提取图像信息，决定每个图像的点是否属于一个图像特征。本文主要探讨如何提取图像中的“角点”这一特征，及其相关的内容。而诸如直方图、边缘、区域等内容在前文中有所提及，请查看相关文章。OpenCv（EmguCv）中实现了多种角点特征的提取方法，包括：Harris角点、ShiTomasi角点、亚像素级角点、SURF角点、Star关键点、FAST关键点、Lepetit关键点等等，本文将逐一介绍如何检测这些角点。在此之前将会先介绍跟角点检测密切相关的一些变换，包括Sobel算子、拉普拉斯算子、Canny算子、霍夫变换。另外，还会介绍一种广泛使用而OpenCv中并未实现的SIFT角点检测，以及最近在OpenCv中实现的MSER区域检测。所要讲述的内容会很多，我这里尽量写一些需要注意的地方及实现代码，而参考手册及书本中有的内容将一笔带过或者不会提及。

Sobel算子

Sobel算子用多项式计算来拟合导数计算，可以用OpenCv中的cvSobel函数或者EmguCv中的Image<TColor,TDepth>.Sobel方法来进行计算。需要注意的是，xorder和yorder中必须且只能有一个为非零值，即只能计算x方向或者y反向的导数；如果将方形滤波器的宽度设置为特殊值CV_SCHARR（-1），将使用Scharr滤波器代替Sobel滤波器。

使用Sobel滤波器的示例代码如下：

//Sobel算子
private string SobelFeatureDetect()
{
//获取参数
int xOrder = int.Parse((string)cmbSobelXOrder.SelectedItem);
int yOrder = int.Parse((string)cmbSobelYOrder.SelectedItem);
int apertureSize = int.Parse((string)cmbSobelApertureSize.SelectedItem);
if ((xOrder == 0 && yOrder == 0) || (xOrder != 0 && yOrder != 0))
return "Sobel算子，参数错误：xOrder和yOrder中必须且只能有一个非零。\r\n";
//计算
Stopwatch sw = new Stopwatch();
sw.Start();
Image<Gray, Single> imageDest = imageSourceGrayscale.Sobel(xOrder, yOrder, apertureSize);
sw.Stop();
//显示
pbResult.Image = imageDest.Bitmap;
//释放资源
imageDest.Dispose();
//返回
return string.Format("·Sobel算子，用时{0:F05}毫秒，参数（x方向求导阶数：{1}，y方向求导阶数：{2}，方形滤波器宽度：{3}）\r\n", sw.Elapsed.TotalMilliseconds, xOrder, yOrder, apertureSize);
}

拉普拉斯算子

拉普拉斯算子可以用作边缘检测；可以用OpenCv中的cvLaplace函数或者EmguCv中的Image<TColor,TDepth>.Laplace方法来进行拉普拉斯变换。需要注意的是：OpenCv的文档有点小错误，apertureSize参数值不能为CV_SCHARR（-1）。

使用拉普拉斯变换的示例代码如下：

//拉普拉斯变换
private string LaplaceFeatureDetect()
{
//获取参数
int apertureSize = int.Parse((string)cmbLaplaceApertureSize.SelectedItem);
//计算
Stopwatch sw = new Stopwatch();
sw.Start();
Image<Gray, Single> imageDest = imageSourceGrayscale.Laplace(apertureSize);
sw.Stop();
//显示
pbResult.Image = imageDest.Bitmap;
//释放资源
imageDest.Dispose();
//返回
return string.Format("·拉普拉斯变换，用时{0:F05}毫秒，参数（方形滤波器宽度：{1}）\r\n", sw.Elapsed.TotalMilliseconds, apertureSize);
}

Canny算子

Canny算子也可以用作边缘检测；可以用OpenCv中的cvCanny函数或者EmguCv中的Image<TColor,TDepth>.Canny方法来进行Canny边缘检测。所不同的是，Image<TColor,TDepth>.Canny方法可以用于检测彩色图像的边缘，但是它只能使用apertureSize参数的默认值3；

而cvCanny只能处理灰度图像，不过可以自定义apertureSize。cvCanny和Canny的方法参数名有点点不同，下面是参数对照表。

Image<TColor,TDepth>.Canny    CvInvoke.cvCanny

thresh                                         lowThresh

threshLinking                               highThresh

3                                                apertureSize

值得注意的是，apertureSize只能取3，5或者7，这可以在cvcanny.cpp第87行看到：

aperture_size &= INT_MAX;

if( (aperture_size & 1) == 0 || aperture_size < 3 || aperture_size > 7 )

CV_ERROR( CV_StsBadFlag, "" );

使用Canny算子的示例代码如下：

//Canny算子
private string CannyFeatureDetect()
{
//获取参数
double lowThresh = double.Parse(txtCannyLowThresh.Text);
double highThresh = double.Parse(txtCannyHighThresh.Text);
int apertureSize = int.Parse((string)cmbCannyApertureSize.SelectedItem);
//计算
Stopwatch sw = new Stopwatch();
sw.Start();
Image<Gray, Byte> imageDest = null;
Image<Bgr, Byte> imageDest2 = null;
if (rbCannyUseCvCanny.Checked)
{
imageDest = new Image<Gray, byte>(imageSourceGrayscale.Size);
CvInvoke.cvCanny(imageSourceGrayscale.Ptr, imageDest.Ptr, lowThresh, highThresh, apertureSize);
}
else
imageDest2 = imageSource.Canny(new Bgr(lowThresh, lowThresh, lowThresh), new Bgr(highThresh, highThresh, highThresh));
sw.Stop();
//显示
pbResult.Image = rbCannyUseCvCanny.Checked ? imageDest.Bitmap : imageDest2.Bitmap;
//释放资源
if (imageDest != null)
imageDest.Dispose();
if (imageDest2 != null)
imageDest2.Dispose();
//返回
return string.Format("·Canny算子，用时{0:F05}毫秒，参数（方式：{1}，阀值下限：{2}，阀值上限：{3}，方形滤波器宽度：{4}）\r\n", sw.Elapsed.TotalMilliseconds, rbCannyUseCvCanny.Checked ? "cvCanny" : "Image<TColor, TDepth>.Canny", lowThresh, highThresh, apertureSize);
}

另外，在http://www.china-vision.net/blog/user2/15975/archives/2007/804.html有一种自动获取Canny算子高低阀值的方法，作者提供了用C语言实现的代码。我将其改写成了C#版本，代码如下：

/// <summary>
/// 计算图像的自适应Canny算子阀值
/// </summary>
/// <param name="imageSrc">源图像，只能是256级灰度图像</param>
/// <param name="apertureSize">方形滤波器的宽度</param>
/// <param name="lowThresh">阀值下限</param>
/// <param name="highThresh">阀值上限</param>
unsafe void AdaptiveFindCannyThreshold(Image<Gray, Byte> imageSrc, int apertureSize, out double lowThresh, out double highThresh)
{
//计算源图像x方向和y方向的1阶Sobel算子
Size size = imageSrc.Size;
Image<Gray, Int16> imageDx = new Image<Gray, short>(size);
Image<Gray, Int16> imageDy = new Image<Gray, short>(size);
CvInvoke.cvSobel(imageSrc.Ptr, imageDx.Ptr, 1, 0, apertureSize);
CvInvoke.cvSobel(imageSrc.Ptr, imageDy.Ptr, 0, 1, apertureSize);
Image<Gray, Single> image = new Image<Gray, float>(size);
int i, j;
DenseHistogram hist = null;
int hist_size = 255;
float[] range_0 = new float[] { 0, 256 };
double PercentOfPixelsNotEdges = 0.7;
//计算边缘的强度，并保存于图像中
float maxv = 0;
float temp;
byte* imageDataDx = (byte*)imageDx.MIplImage.imageData.ToPointer();
byte* imageDataDy = (byte*)imageDy.MIplImage.imageData.ToPointer();
byte* imageData = (byte*)image.MIplImage.imageData.ToPointer();
int widthStepDx = imageDx.MIplImage.widthStep;
int widthStepDy = widthStepDx;
int widthStep = image.MIplImage.widthStep;
for (i = 0; i < size.Height; i++)
{
short* _dx = (short*)(imageDataDx + widthStepDx * i);
short* _dy = (short*)(imageDataDy + widthStepDy * i);
float* _image = (float*)(imageData + widthStep * i);
for (j = 0; j < size.Width; j++)
{
    temp = (float)(Math.Abs(*(_dx + j)) + Math.Abs(*(_dy + j)));
    *(_image + j) = temp;
    if (maxv < temp)
maxv = temp;
}
}
//计算直方图
range_0[1] = maxv;
hist_size = hist_size > maxv ? (int)maxv : hist_size;
hist = new DenseHistogram(hist_size, new RangeF(range_0[0], range_0[1]));
hist.Calculate<Single>(new Image<Gray, Single>[] { image }, false, null);
int total = (int)(size.Height * size.Width * PercentOfPixelsNotEdges);
double sum = 0;
int icount = hist.BinDimension[0].Size;
for (i = 0; i < icount; i++)
{
sum += hist[i];
if (sum > total)
    break;
}
//计算阀值
highThresh = (i + 1) * maxv / hist_size;
lowThresh = highThresh * 0.4;
//释放资源
imageDx.Dispose();
imageDy.Dispose(); image.Dispose();
hist.Dispose();
}

霍夫变换

霍夫变换是一种在图像中寻找直线、圆及其他简单形状的方法，在OpenCv中实现了霍夫线变换和霍夫圆变换。值得注意的地方有以下几点：（1）HoughLines2需要先计算Canny边缘，然后再检测直线；（2）HoughLines2计算结果的获取随获取方式的不同而不同；（3）HoughCircles检测结果似乎不正确。

使用霍夫变换的示例代码如下所示：

//霍夫线变换
private string HoughLinesFeatureDetect()
{
//获取参数
HOUGH_TYPE method = rbHoughLinesSHT.Checked ? HOUGH_TYPE.CV_HOUGH_STANDARD : (rbHoughLinesPPHT.Checked ? HOUGH_TYPE.CV_HOUGH_PROBABILISTIC : HOUGH_TYPE.CV_HOUGH_MULTI_SCALE);
double rho = double.Parse(txtHoughLinesRho.Text);
double theta = double.Parse(txtHoughLinesTheta.Text);
int threshold = int.Parse(txtHoughLinesThreshold.Text);
double param1 = double.Parse(txtHoughLinesParam1.Text);
double param2 = double.Parse(txtHoughLinesParam2.Text);
MemStorage storage = new MemStorage();
int linesCount = 0;
StringBuilder sbResult = new StringBuilder();
//计算，先运行Canny边缘检测（参数来自Canny算子属性页），然后再用计算霍夫线变换
double lowThresh = double.Parse(txtCannyLowThresh.Text);
double highThresh = double.Parse(txtCannyHighThresh.Text);
int apertureSize = int.Parse((string)cmbCannyApertureSize.SelectedItem);
Image<Gray, Byte> imageCanny = new Image<Gray, byte>(imageSourceGrayscale.Size);
CvInvoke.cvCanny(imageSourceGrayscale.Ptr, imageCanny.Ptr, lowThresh, highThresh, apertureSize);
Stopwatch sw = new Stopwatch();
sw.Start();
IntPtr ptrLines = CvInvoke.cvHoughLines2(imageCanny.Ptr, storage.Ptr, method, rho, theta, threshold, param1, param2);
Seq<LineSegment2D> linesSeq = null;
Seq<PointF> linesSeq2 = null;
if (method == HOUGH_TYPE.CV_HOUGH_PROBABILISTIC)
linesSeq = new Seq<LineSegment2D>(ptrLines, storage);
else
linesSeq2 = new Seq<PointF>(ptrLines, storage);
sw.Stop();
//显示
Image<Bgr, Byte> imageResult = imageSourceGrayscale.Convert<Bgr, Byte>();
if (linesSeq != null)
{
linesCount = linesSeq.Total;
foreach (LineSegment2D line in linesSeq)
{
    imageResult.Draw(line, new Bgr(255d, 0d, 0d), 4);
    sbResult.AppendFormat("{0}-{1}，", line.P1, line.P2);
}
}
else
{
linesCount = linesSeq2.Total;
foreach (PointF line in linesSeq2)
{
    float r = line.X;
    float t = line.Y;
    double a = Math.Cos(t), b = Math.Sin(t);
    double x0 = a * r, y0 = b * r;
    int x1 = (int)(x0 + 1000 * (-b));
    int y1 = (int)(y0 + 1000 * (a));
    int x2 = (int)(x0 - 1000 * (-b));
    int y2 = (int)(y0 - 1000 * (a));
    Point pt1 = new Point(x1, y1);
    Point pt2 = new Point(x2, y2);
    imageResult.Draw(new LineSegment2D(pt1, pt2), new Bgr(255d, 0d, 0d), 4);
    sbResult.AppendFormat("{0}-{1}，", pt1, pt2);
}
}
pbResult.Image = imageResult.Bitmap;
//释放资源
imageCanny.Dispose();
imageResult.Dispose();
storage.Dispose();
//返回
return string.Format("·霍夫线变换，用时{0:F05}毫秒，参数（变换方式：{1}，距离精度：{2}，弧度精度：{3}，阀值：{4}，参数1：{5}，参数2：{6}），找到{7}条直线\r\n{8}",
sw.Elapsed.TotalMilliseconds, method.ToString("G"), rho, theta, threshold, param1, param2, linesCount, linesCount != 0 ? (sbResult.ToString() + "\r\n") : "");
}

//霍夫圆变换
private string HoughCirclesFeatureDetect()
{
//获取参数
double dp = double.Parse(txtHoughCirclesDp.Text);
double minDist = double.Parse(txtHoughCirclesMinDist.Text);
double param1 = double.Parse(txtHoughCirclesParam1.Text);
double param2 = double.Parse(txtHoughCirclesParam2.Text);
int minRadius = int.Parse(txtHoughCirclesMinRadius.Text);
int maxRadius = int.Parse(txtHoughCirclesMaxRadius.Text);
StringBuilder sbResult = new StringBuilder();
//计算
Stopwatch sw = new Stopwatch();
sw.Start();
CircleF[][] circles = imageSourceGrayscale.HoughCircles(new Gray(param1), new Gray(param2), dp, minDist, minRadius, maxRadius);
sw.Stop();
//显示
Image<Bgr, Byte> imageResult = imageSourceGrayscale.Convert<Bgr, Byte>();
int circlesCount = 0;
foreach (CircleF[] cs in circles)
{
foreach (CircleF circle in cs)
{
    imageResult.Draw(circle, new Bgr(255d, 0d, 0d), 4);
    sbResult.AppendFormat("圆心{0}半径{1}，", circle.Center, circle.Radius);
    circlesCount++;
}
}
pbResult.Image = imageResult.Bitmap;
//释放资源
imageResult.Dispose();
//返回
return string.Format("·霍夫圆变换，用时{0:F05}毫秒，参数（累加器图像的最小分辨率：{1}，不同圆之间的最小距离：{2}，边缘阀值：{3}，累加器阀值：{4}，最小圆半径：{5}，最大圆半径：{6}），找到{7}个圆\r\n{8}",
sw.Elapsed.TotalMilliseconds, dp, minDist, param1, param2, minRadius, maxRadius, circlesCount, sbResult.Length > 0 ? (sbResult.ToString() + "\r\n") : "");
}

Harris角点

cvCornerHarris函数检测的结果实际上是一幅包含Harris角点的浮点型单通道图像，可以使用类似下面的代码来计算包含Harris角点的图像：

//Harris角点
private string CornerHarrisFeatureDetect()
{
//获取参数
int blockSize = int.Parse(txtCornerHarrisBlockSize.Text);
int apertureSize = int.Parse(txtCornerHarrisApertureSize.Text);
double k = double.Parse(txtCornerHarrisK.Text);
//计算
Image<Gray, Single> imageDest = new Image<Gray, float>(imageSourceGrayscale.Size);
Stopwatch sw = new Stopwatch();
sw.Start();
CvInvoke.cvCornerHarris(imageSourceGrayscale.Ptr, imageDest.Ptr, blockSize, apertureSize, k);
sw.Stop();
//显示
pbResult.Image = imageDest.Bitmap;
//释放资源
imageDest.Dispose();
//返回
return string.Format("·Harris角点，用时{0:F05}毫秒，参数（邻域大小：{1}，方形滤波器宽度：{2}，权重系数：{3}）\r\n", sw.Elapsed.TotalMilliseconds, blockSize, apertureSize, k);
}

如果要计算Harris角点列表，需要使用cvGoodFeatureToTrack函数，并传递适当的参数。

ShiTomasi角点

在默认情况下，cvGoodFeatureToTrack函数计算ShiTomasi角点；不过如果将参数use_harris设置为非0值，那么它会计算harris角点。

使用cvGoodFeatureToTrack函数的示例代码如下：

//ShiTomasi角点
private string CornerShiTomasiFeatureDetect()
{
//获取参数
int cornerCount = int.Parse(txtGoodFeaturesCornerCount.Text);
double qualityLevel = double.Parse(txtGoodFeaturesQualityLevel.Text);
double minDistance = double.Parse(txtGoodFeaturesMinDistance.Text);
int blockSize = int.Parse(txtGoodFeaturesBlockSize.Text);
bool useHarris = cbGoodFeaturesUseHarris.Checked;
double k = double.Parse(txtGoodFeaturesK.Text);
//计算
Stopwatch sw = new Stopwatch();
sw.Start();
PointF[][] corners = imageSourceGrayscale.GoodFeaturesToTrack(cornerCount, qualityLevel, minDistance, blockSize, useHarris, k);
sw.Stop();
//显示
Image<Bgr, Byte> imageResult = imageSourceGrayscale.Convert<Bgr, Byte>();
int cornerCount2 = 0;
StringBuilder sbResult = new StringBuilder();
int radius = (int)(minDistance / 2) + 1;
int thickness = (int)(minDistance / 4) + 1;
foreach (PointF[] cs in corners)
{
foreach (PointF p in cs)
{
    imageResult.Draw(new CircleF(p, radius), new Bgr(255d, 0d, 0d), thickness);
    cornerCount2++;
    sbResult.AppendFormat("{0}，", p);
}
}
pbResult.Image = imageResult.Bitmap;
//释放资源
imageResult.Dispose();
//返回
return string.Format("·ShiTomasi角点，用时{0:F05}毫秒，参数（最大角点数目：{1}，最小特征值：{2}，角点间的最小距离：{3}，邻域大小：{4}，角点类型：{5}，权重系数：{6}），检测到{7}个角点\r\n{8}",
sw.Elapsed.TotalMilliseconds, cornerCount, qualityLevel, minDistance, blockSize, useHarris ? "Harris" : "ShiTomasi", k, cornerCount2, cornerCount2 > 0 ? (sbResult.ToString() + "\r\n") : "");
}

亚像素级角点

在检测亚像素级角点前，需要提供角点的初始为止，这些初始位置可以用本文给出的其他的角点检测方式来获取，不过使用GoodFeaturesToTrack得到的结果最方便直接使用。

亚像素级角点检测的示例代码如下：

//亚像素级角点
private string CornerSubPixFeatureDetect()
{
//获取参数
int winWidth = int.Parse(txtCornerSubPixWinWidth.Text);
int winHeight = int.Parse(txtCornerSubPixWinHeight.Text);
Size win = new Size(winWidth, winHeight);
int zeroZoneWidth = int.Parse(txtCornerSubPixZeroZoneWidth.Text);
int zeroZoneHeight = int.Parse(txtCornerSubPixZeroZoneHeight.Text);
Size zeroZone = new Size(zeroZoneWidth, zeroZoneHeight);
int maxIter=int.Parse(txtCornerSubPixMaxIter.Text);
double epsilon=double.Parse(txtCornerSubPixEpsilon.Text);
MCvTermCriteria criteria = new MCvTermCriteria(maxIter, epsilon);
//先计算得到易于跟踪的点（ShiTomasi角点）
int cornerCount = int.Parse(txtGoodFeaturesCornerCount.Text);
double qualityLevel = double.Parse(txtGoodFeaturesQualityLevel.Text);
double minDistance = double.Parse(txtGoodFeaturesMinDistance.Text);
int blockSize = int.Parse(txtGoodFeaturesBlockSize.Text);
bool useHarris = cbGoodFeaturesUseHarris.Checked;
double k = double.Parse(txtGoodFeaturesK.Text);
PointF[][] corners = imageSourceGrayscale.GoodFeaturesToTrack(cornerCount, qualityLevel, minDistance, blockSize, useHarris, k);
//计算
Stopwatch sw = new Stopwatch();
sw.Start();
imageSourceGrayscale.FindCornerSubPix(corners, win, zeroZone, criteria);
sw.Stop();
//显示
Image<Bgr, Byte> imageResult = imageSourceGrayscale.Convert<Bgr, Byte>();
int cornerCount2 = 0;
StringBuilder sbResult = new StringBuilder();
int radius = (int)(minDistance / 2) + 1;
int thickness = (int)(minDistance / 4) + 1;
foreach (PointF[] cs in corners)
{
foreach (PointF p in cs)
{
    imageResult.Draw(new CircleF(p, radius), new Bgr(255d, 0d, 0d), thickness);
    cornerCount2++;
    sbResult.AppendFormat("{0}，", p);
}
}
pbResult.Image = imageResult.Bitmap;
//释放资源
imageResult.Dispose();
//返回
return string.Format("·亚像素级角点，用时{0:F05}毫秒，参数（搜索窗口：{1}，死区：{2}，最大迭代次数：{3}，亚像素值的精度：{4}），检测到{5}个角点\r\n{6}",
sw.Elapsed.TotalMilliseconds, win, zeroZone, maxIter, epsilon, cornerCount2, cornerCount2 > 0 ? (sbResult.ToString() + "\r\n") : "");
}

SURF角点

OpenCv中的cvExtractSURF函数和EmguCv中的Image<TColor,TDepth>.ExtractSURF方法用于检测SURF角点。

SURF角点检测的示例代码如下：

//SURF角点
private string SurfFeatureDetect()
{
//获取参数
bool getDescriptors = cbSurfGetDescriptors.Checked;
MCvSURFParams surfParam = new MCvSURFParams();
surfParam.extended=rbSurfBasicDescriptor.Checked ? 0 : 1;
surfParam.hessianThreshold=double.Parse(txtSurfHessianThreshold.Text);
surfParam.nOctaves=int.Parse(txtSurfNumberOfOctaves.Text);
surfParam.nOctaveLayers=int.Parse(txtSurfNumberOfOctaveLayers.Text);
//计算
SURFFeature[] features = null;
MKeyPoint[] keyPoints = null;
Stopwatch sw = new Stopwatch();
sw.Start();
if (getDescriptors)
features = imageSourceGrayscale.ExtractSURF(ref surfParam);
else
keyPoints = surfParam.DetectKeyPoints(imageSourceGrayscale, null);
sw.Stop();
//显示
bool showDetail = cbSurfShowDetail.Checked;
Image<Bgr, Byte> imageResult = imageSourceGrayscale.Convert<Bgr, Byte>();
StringBuilder sbResult = new StringBuilder();
int idx = 0;
if (getDescriptors)
{
foreach (SURFFeature feature in features)
{
    imageResult.Draw(new CircleF(feature.Point.pt, 5), new Bgr(255d, 0d, 0d), 2);
    if (showDetail)
    {
sbResult.AppendFormat("第{0}点（坐标：{1}，尺寸：{2}，方向：{3}°，hessian值：{4}，拉普拉斯标志：{5}，描述：[",
    idx, feature.Point.pt, feature.Point.size, feature.Point.dir, feature.Point.hessian, feature.Point.laplacian);
foreach (float d in feature.Descriptor)
    sbResult.AppendFormat("{0},", d);
sbResult.Append("]），");
    }
    idx++;
}
}
else
{
foreach (MKeyPoint keypoint in keyPoints)
{
    imageResult.Draw(new CircleF(keypoint.Point, 5), new Bgr(255d, 0d, 0d), 2);
    if (showDetail)
sbResult.AppendFormat("第{0}点（坐标：{1}，尺寸：{2}，方向：{3}°，响应：{4}，octave：{5}），",
    idx, keypoint.Point, keypoint.Size, keypoint.Angle, keypoint.Response, keypoint.Octave);
    idx++;
}
}
pbResult.Image = imageResult.Bitmap;
//释放资源
imageResult.Dispose();
//返回
return string.Format("·SURF角点，用时{0:F05}毫秒，参数（描述：{1}，hessian阀值：{2}，octave数目：{3}，每个octave的层数：{4}，检测到{5}个角点\r\n{6}",
sw.Elapsed.TotalMilliseconds, getDescriptors ? (surfParam.extended == 0 ? "获取基本描述" : "获取扩展描述") : "不获取描述", surfParam.hessianThreshold,
surfParam.nOctaves, surfParam.nOctaveLayers, getDescriptors ? features.Length : keyPoints.Length, showDetail ? sbResult.ToString() + "\r\n" : "");
}

Star关键点

OpenCv中的cvGetStarKeypoints函数和EmguCv中的Image<TColor,TDepth>.GetStarKeypoints方法用于检测“星型”附近的点。

Star关键点检测的示例代码如下：

//Star关键点
private string StarKeyPointFeatureDetect()
{
//获取参数
StarDetector starParam = new StarDetector();
starParam.MaxSize = int.Parse((string)cmbStarMaxSize.SelectedItem);
starParam.ResponseThreshold = int.Parse(txtStarResponseThreshold.Text);
starParam.LineThresholdProjected = int.Parse(txtStarLineThresholdProjected.Text);
starParam.LineThresholdBinarized = int.Parse(txtStarLineThresholdBinarized.Text);
starParam.SuppressNonmaxSize = int.Parse(txtStarSuppressNonmaxSize.Text);
//计算
Stopwatch sw = new Stopwatch();
sw.Start();
MCvStarKeypoint[] keyPoints = imageSourceGrayscale.GetStarKeypoints(ref starParam);
sw.Stop();
//显示
Image<Bgr, Byte> imageResult = imageSourceGrayscale.Convert<Bgr, Byte>();
StringBuilder sbResult = new StringBuilder();
int idx = 0;
foreach (MCvStarKeypoint keypoint in keyPoints)
{
imageResult.Draw(new CircleF(new PointF(keypoint.pt.X, keypoint.pt.Y), keypoint.size / 2), new Bgr(255d, 0d, 0d), keypoint.size / 4);
sbResult.AppendFormat("第{0}点（坐标：{1}，尺寸：{2}，强度：{3}），", idx, keypoint.pt, keypoint.size, keypoint.response);
idx++;
}
pbResult.Image = imageResult.Bitmap;
//释放资源
imageResult.Dispose();
//返回
return string.Format("·Star关键点，用时{0:F05}毫秒，参数（MaxSize：{1}，ResponseThreshold：{2}，LineThresholdProjected：{3}，LineThresholdBinarized：{4}，SuppressNonmaxSize：{5}），检测到{6}个关键点\r\n{7}",
sw.Elapsed.TotalMilliseconds, starParam.MaxSize, starParam.ResponseThreshold, starParam.LineThresholdProjected, starParam.LineThresholdBinarized, starParam.SuppressNonmaxSize, keyPoints.Length, keyPoints.Length > 0 ? (sbResult.ToString() + "\r\n") : "");
}

FAST角点检测

FAST角点由E. Rosten教授提出，相比其他检测手段，这种方法的速度正如其名，相当的快。值得关注的是他所研究的理论都是属于实用类的，都很快。Rosten教授实现了FAST角点检测，并将其提供给了OpenCv，相当的有爱呀；不过OpenCv中的函数和Rosten教授的实现似乎有点点不太一样。遗憾的是，OpenCv中目前还没有FAST角点检测的文档。下面是我从Rosten的代码中找到的函数声明，可以看到粗略的参数说明。

/*

The references are:

* Machine learning for high-speed corner detection,

E. Rosten and T. Drummond, ECCV 2006

* Faster and better: A machine learning approach to corner detection

E. Rosten, R. Porter and T. Drummond, PAMI, 2009

*/

void cvCornerFast( const CvArr* image, int threshold, int N,

   int nonmax_suppression, int* ret_number_of_corners,

   CvPoint** ret_corners);

image:      OpenCV image in which to detect corners. Must be 8 bit unsigned.

threshold:  Threshold for detection (higher is fewer corners). 0--255

N:          Arc length of detector, 9, 10, 11 or 12. 9 is usually best.

nonmax_suppression: Whether to perform nonmaximal suppression.

ret_number_of_corners: The number of detected corners is returned here.

ret_corners: The corners are returned here.

EmguCv中的Image<TColor,TDepth>.GetFASTKeypoints方法也实现了FAST角点检测，不过参数少了一些，只有threshold和nonmaxSupression，其中N我估计取的默认值9，但是返回的角点数目我不知道是怎么设置的。

使用FAST角点检测的示例代码如下：

//FAST关键点
private string FASTKeyPointFeatureDetect()
{
//获取参数
int threshold = int.Parse(txtFASTThreshold.Text);
bool nonmaxSuppression = cbFASTNonmaxSuppression.Checked;
bool showDetail = cbFASTShowDetail.Checked;
//计算
Stopwatch sw = new Stopwatch();
sw.Start();
MKeyPoint[] keyPoints = imageSourceGrayscale.GetFASTKeypoints(threshold, nonmaxSuppression);
sw.Stop();
//显示
Image<Bgr, Byte> imageResult = imageSourceGrayscale.Convert<Bgr, Byte>();
StringBuilder sbResult = new StringBuilder();
int idx = 0;
foreach (MKeyPoint keypoint in keyPoints)
{
imageResult.Draw(new CircleF(keypoint.Point, (int)(keypoint.Size / 2)), new Bgr(255d, 0d, 0d), (int)(keypoint.Size / 4));
if (showDetail)
    sbResult.AppendFormat("第{0}点（坐标：{1}，尺寸：{2}，方向：{3}°，响应：{4}，octave：{5}），",
idx, keypoint.Point, keypoint.Size, keypoint.Angle, keypoint.Response, keypoint.Octave);
idx++;
}
pbResult.Image = imageResult.Bitmap;
//释放资源
imageResult.Dispose();
//返回
return string.Format("·FAST关键点，用时{0:F05}毫秒，参数（阀值：{1}，nonmaxSupression：{2}），检测到{3}个关键点\r\n{4}",
sw.Elapsed.TotalMilliseconds, threshold, nonmaxSuppression, keyPoints.Length, showDetail ? (sbResult.ToString() + "\r\n") : "");
}

Lepetit关键点

Lepetit关键点由Vincent Lepetit提出，可以在他的网站（http://cvlab.epfl.ch/~vlepetit/）上看到相关的论文等资料。EmguCv中的类LDetector实现了Lepetit关键点的检测。

使用Lepetit关键点检测的示例代码如下：

//Lepetit关键点
private string LepetitKeyPointFeatureDetect()
{
//获取参数
LDetector lepetitDetector = new LDetector();
lepetitDetector.BaseFeatureSize = double.Parse(txtLepetitBaseFeatureSize.Text);
lepetitDetector.ClusteringDistance = double.Parse(txtLepetitClasteringDistance.Text);
lepetitDetector.NOctaves = int.Parse(txtLepetitNumberOfOctaves.Text);
lepetitDetector.NViews = int.Parse(txtLepetitNumberOfViews.Text);
lepetitDetector.Radius = int.Parse(txtLepetitRadius.Text);
lepetitDetector.Threshold = int.Parse(txtLepetitThreshold.Text);
lepetitDetector.Verbose = cbLepetitVerbose.Checked;
int maxCount = int.Parse(txtLepetitMaxCount.Text);
bool scaleCoords = cbLepetitScaleCoords.Checked;
bool showDetail = cbLepetitShowDetail.Checked;
//计算
Stopwatch sw = new Stopwatch();
sw.Start();
MKeyPoint[] keyPoints = lepetitDetector.DetectKeyPoints(imageSourceGrayscale, maxCount, scaleCoords);
sw.Stop();
//显示
Image<Bgr, Byte> imageResult = imageSourceGrayscale.Convert<Bgr, Byte>();
StringBuilder sbResult = new StringBuilder();
int idx = 0;
foreach (MKeyPoint keypoint in keyPoints)
{
//imageResult.Draw(new CircleF(keypoint.Point, (int)(keypoint.Size / 2)), new Bgr(255d, 0d, 0d), (int)(keypoint.Size / 4));
imageResult.Draw(new CircleF(keypoint.Point, 4), new Bgr(255d, 0d, 0d), 2);
if (showDetail)
    sbResult.AppendFormat("第{0}点（坐标：{1}，尺寸：{2}，方向：{3}°，响应：{4}，octave：{5}），",
idx, keypoint.Point, keypoint.Size, keypoint.Angle, keypoint.Response, keypoint.Octave);
idx++;
}
pbResult.Image = imageResult.Bitmap;
//释放资源
imageResult.Dispose();
//返回
return string.Format("·Lepetit关键点，用时{0:F05}毫秒，参数（基础特征尺寸：{1}，集群距离：{2}，阶数：{3}，视图数：{4}，半径：{5}，阀值：{6}，计算详细结果：{7}，最大关键点数目：{8}，缩放坐标：{9}），检测到{10}个关键点\r\n{11}",
sw.Elapsed.TotalMilliseconds, lepetitDetector.BaseFeatureSize, lepetitDetector.ClusteringDistance, lepetitDetector.NOctaves, lepetitDetector.NViews,
lepetitDetector.Radius, lepetitDetector.Threshold, lepetitDetector.Verbose, maxCount, scaleCoords, keyPoints.Length, showDetail ? (sbResult.ToString() + "\r\n") : "");
}

SIFT角点

SIFT角点是一种广泛使用的图像特征，可用于物体跟踪、图像匹配、图像拼接等领域，然而奇怪的是它并未被OpenCv实现。提出SIFT角点的David Lowe教授已经用C和matlab实现了SIFT角点的检测，并开放了源代码，不过他的实现不方便直接使用。您可以在http://www.cs.ubc.ca/~lowe/keypoints/看到SIFT的介绍、相关论文及David Lowe教授的实现代码。下面我要介绍由Andrea Vedaldi和Brian Fulkerson先生创建的vlfeat开源图像处理库，vlfeat库有C和matlab两种实现，其中包含了SIFT检测。您可以在http://www.vlfeat.org/下载到vlfeat库的代码、文档及可执行文件。

使用vlfeat检测SIFT角点需要以下步骤：

（1）用函数vl_sift_new()初始化SIFT过滤器对象，该过滤器对象可以反复用于多幅尺寸相同的图像；

（2）用函数vl_sift_first_octave()及vl_sift_process_next()遍历缩放空间的每一阶，直到返回VL_ERR_EOF为止；

（3）对于缩放空间的每一阶，用函数vl_sift_detect()来获取关键点；

（4）对每个关键点，用函数vl_sift_calc_keypoint_orientations()来获取该点的方向；

（5）对关键点的每个方向，用函数vl_sift_calc_keypoint_descriptor()来获取该方向的描述；

（6）使用完之后，用函数vl_sift_delete()来释放资源；

（7）如果要计算某个自定义关键点的描述，可以使用函数vl_sift_calc_raw_descriptor()。

直接使用vlfeat中的SIFT角点检测示例代码如下：

//通过P/Invoke调用vlfeat函数来进行SIFT检测
unsafe private string SiftFeatureDetectByPinvoke(int noctaves, int nlevels, int o_min, bool showDetail)
{
StringBuilder sbResult = new StringBuilder();
//初始化
IntPtr ptrSiftFilt = VlFeatInvoke.vl_sift_new(imageSource.Width, imageSource.Height, noctaves, nlevels, o_min);
if (ptrSiftFilt == IntPtr.Zero)
return "Sift特征检测：初始化失败。";
//处理
Image<Gray, Single> imageSourceSingle = imageSourceGrayscale.ConvertScale<Single>(1d, 0d);
Image<Bgr, Byte> imageResult = imageSourceGrayscale.Convert<Bgr, Byte>();
int pointCount = 0;
int idx = 0;
//依次遍历每一组
if (VlFeatInvoke.vl_sift_process_first_octave(ptrSiftFilt, imageSourceSingle.MIplImage.imageData) != VlFeatInvoke.VL_ERR_EOF)
{
while (true)
{
    //计算每组中的关键点
    VlFeatInvoke.vl_sift_detect(ptrSiftFilt);
    //遍历并绘制每个点
    VlSiftFilt siftFilt = (VlSiftFilt)Marshal.PtrToStructure(ptrSiftFilt, typeof(VlSiftFilt));
    pointCount += siftFilt.nkeys;
    VlSiftKeypoint* pKeyPoints = (VlSiftKeypoint*)siftFilt.keys.ToPointer();
    for (int i = 0; i < siftFilt.nkeys; i++)
    {
VlSiftKeypoint keyPoint = *pKeyPoints;
pKeyPoints++;
imageResult.Draw(new CircleF(new PointF(keyPoint.x, keyPoint.y), keyPoint.sigma / 2), new Bgr(255d, 0d, 0d), 2);
if (showDetail)
    sbResult.AppendFormat("第{0}点，坐标：({1},{2})，阶：{3}，缩放：{4}，s：{5}，", idx, keyPoint.x, keyPoint.y, keyPoint.o, keyPoint.sigma, keyPoint.s);
idx++;
//计算并遍历每个点的方向
double[] angles = new double[4];
int angleCount = VlFeatInvoke.vl_sift_calc_keypoint_orientations(ptrSiftFilt, angles, ref keyPoint);
if (showDetail)
    sbResult.AppendFormat("共{0}个方向，", angleCount);
for (int j = 0; j < angleCount; j++)
{
    double angle = angles[j];
    if (showDetail)
sbResult.AppendFormat("【方向：{0}，描述：", angle);
    //计算每个方向的描述
    IntPtr ptrDescriptors = Marshal.AllocHGlobal(128 * sizeof(float));
    VlFeatInvoke.vl_sift_calc_keypoint_descriptor(ptrSiftFilt, ptrDescriptors, ref keyPoint, angle);
    float* pDescriptors = (float*)ptrDescriptors.ToPointer();
    for (int k = 0; k < 128; k++)
    {
float descriptor = *pDescriptors;
pDescriptors++;
if (showDetail)
    sbResult.AppendFormat("{0},", descriptor);
    }
    sbResult.Append("】，");
    Marshal.FreeHGlobal(ptrDescriptors);
}
    }
    //下一阶
    if (VlFeatInvoke.vl_sift_process_next_octave(ptrSiftFilt) == VlFeatInvoke.VL_ERR_EOF)
break;
}
}
//显示
pbResult.Image = imageResult.Bitmap;
//释放资源
VlFeatInvoke.vl_sift_delete(ptrSiftFilt);
imageSourceSingle.Dispose();
imageResult.Dispose();
//返回
return string.Format("·SIFT特征检测(P/Invoke)，用时：未统计，参数（阶数：{0}，每阶层数：{1}，最小阶索引：{2}），{3}个关键点\r\n{4}",
noctaves, nlevels, o_min, pointCount, showDetail ? (sbResult.ToString() + "\r\n") : "");
}

要在.net中使用vlfeat还是不够方便，为此我对vlfeat中的SIFT角点检测部分进行了封装，将相关操作放到了类SiftDetector中。

使用SiftDetector需要两至三步：

（1）用构造函数初始化SiftDetector对象；

（2）用Process方法计算特征；

（3）视需要调用Dispose方法释放资源，或者等待垃圾回收器来自动释放资源。

使用SiftDetector的示例代码如下：

//通过dotnet封装的SiftDetector类来进行SIFT检测
private string SiftFeatureDetectByDotNet(int noctaves, int nlevels, int o_min, bool showDetail)
{
//初始化对象
SiftDetector siftDetector = new SiftDetector(imageSource.Size, noctaves, nlevels, o_min);
//计算
Image<Gray, Single> imageSourceSingle = imageSourceGrayscale.Convert<Gray, Single>();
Stopwatch sw = new Stopwatch();
sw.Start();
List<SiftFeature> features = siftDetector.Process(imageSourceSingle, showDetail ? SiftDetectorResultType.Extended : SiftDetectorResultType.Basic);
sw.Stop();
//显示结果
Image<Bgr, Byte> imageResult = imageSourceGrayscale.Convert<Bgr, Byte>();
StringBuilder sbResult = new StringBuilder();
int idx=0;
foreach (SiftFeature feature in features)
{
imageResult.Draw(new CircleF(new PointF(feature.keypoint.x, feature.keypoint.y), feature.keypoint.sigma / 2), new Bgr(255d, 0d, 0d), 2);
if (showDetail)
{
    sbResult.AppendFormat("第{0}点，坐标：({1},{2})，阶：{3}，缩放：{4}，s：{5}，",
idx, feature.keypoint.x, feature.keypoint.y, feature.keypoint.o, feature.keypoint.sigma, feature.keypoint.s);
    sbResult.AppendFormat("共{0}个方向，", feature.keypointOrientations != null ? feature.keypointOrientations.Length : 0);
    if (feature.keypointOrientations != null)
    {
foreach (SiftKeyPointOrientation orientation in feature.keypointOrientations)
{
    if (orientation.descriptors != null)
    {
sbResult.AppendFormat("【方向：{0}，描述：", orientation.angle);
foreach (float descriptor in orientation.descriptors)
    sbResult.AppendFormat("{0},", descriptor);
    }
    else
sbResult.AppendFormat("【方向：{0}，", orientation.angle);
    sbResult.Append("】，");
}
    }
}
}
pbResult.Image = imageResult.Bitmap;
//释放资源
siftDetector.Dispose();
imageSourceSingle.Dispose();
imageResult.Dispose();
//返回
return string.Format("·SIFT特征检测(.net)，用时：{0:F05}毫秒，参数（阶数：{1}，每阶层数：{2}，最小阶索引：{3}），{4}个关键点\r\n{5}",
sw.Elapsed.TotalMilliseconds, noctaves, nlevels, o_min, features.Count, showDetail ? (sbResult.ToString() + "\r\n") : "");
}

对vlfeat库中的SIFT部分封装代码如下所示：

using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Runtime.InteropServices;

namespace ImageProcessLearn
{
[StructLayoutAttribute(LayoutKind.Sequential)]
public struct VlSiftKeypoint
{

/// int
public int o;

/// int
public int ix;

/// int
public int iy;

/// int
public int @is;

/// float
public float x;

/// float
public float y;

/// float
public float s;

/// float
public float sigma;
}

[StructLayoutAttribute(LayoutKind.Sequential)]
public struct VlSiftFilt
{

/// double
public double sigman;

/// double
public double sigma0;

/// double
public double sigmak;

/// double
public double dsigma0;

/// int
public int width;

/// int
public int height;

/// int
public int O;

/// int
public int S;

/// int
public int o_min;

/// int
public int s_min;

/// int
public int s_max;

/// int
public int o_cur;

/// vl_sift_pix*
public System.IntPtr temp;

/// vl_sift_pix*
public System.IntPtr octave;

/// vl_sift_pix*
public System.IntPtr dog;

/// int
public int octave_width;

/// int
public int octave_height;

/// VlSiftKeypoint*
public System.IntPtr keys;

/// int
public int nkeys;

/// int
public int keys_res;

/// double
public double peak_thresh;

/// double
public double edge_thresh;

/// double
public double norm_thresh;

/// double
public double magnif;

/// double
public double windowSize;

/// vl_sift_pix*
public System.IntPtr grad;

/// int
public int grad_o;

/// <summary>
/// 获取SiftFilt指针；
/// 注意在使用完指针之后，需要用Marshal.FreeHGlobal释放内存。
/// </summary>
/// <returns></returns>
unsafe public IntPtr GetPtrOfVlSiftFilt()
{
IntPtr ptrSiftFilt = Marshal.AllocHGlobal(sizeof(VlSiftFilt));
Marshal.StructureToPtr(this, ptrSiftFilt, true);
return ptrSiftFilt;
}
}

public class VlFeatInvoke
{
/// VL_ERR_MSG_LEN -> 1024
public const int VL_ERR_MSG_LEN = 1024;

/// VL_ERR_OK -> 0
public const int VL_ERR_OK = 0;

/// VL_ERR_OVERFLOW -> 1
public const int VL_ERR_OVERFLOW = 1;

/// VL_ERR_ALLOC -> 2
public const int VL_ERR_ALLOC = 2;

/// VL_ERR_BAD_ARG -> 3
public const int VL_ERR_BAD_ARG = 3;

/// VL_ERR_IO -> 4
public const int VL_ERR_IO = 4;

/// VL_ERR_EOF -> 5
public const int VL_ERR_EOF = 5;

/// VL_ERR_NO_MORE -> 5
public const int VL_ERR_NO_MORE = 5;

/// Return Type: VlSiftFilt*
/// int
///height: int
///noctaves: int
///nlevels: int
///o_min: int
[DllImportAttribute("vl.dll", EntryPoint = "vl_sift_new")]
public static extern System.IntPtr vl_sift_new(int width, int height, int noctaves, int nlevels, int o_min);


/// Return Type: void
///f: VlSiftFilt*
[DllImportAttribute("vl.dll", EntryPoint = "vl_sift_delete")]
public static extern void vl_sift_delete(IntPtr f);


/// Return Type: int
///f: VlSiftFilt*
///im: vl_sift_pix*
[DllImportAttribute("vl.dll", EntryPoint = "vl_sift_process_first_octave")]
public static extern int vl_sift_process_first_octave(IntPtr f, IntPtr im);


/// Return Type: int
///f: VlSiftFilt*
[DllImportAttribute("vl.dll", EntryPoint = "vl_sift_process_next_octave")]
public static extern int vl_sift_process_next_octave(IntPtr f);


/// Return Type: void
///f: VlSiftFilt*
[DllImportAttribute("vl.dll", EntryPoint = "vl_sift_detect")]
public static extern void vl_sift_detect(IntPtr f);


/// Return Type: int
///f: VlSiftFilt*
///angles: double*
///k: VlSiftKeypoint*
[DllImportAttribute("vl.dll", EntryPoint = "vl_sift_calc_keypoint_orientations")]
public static extern int vl_sift_calc_keypoint_orientations(IntPtr f, double[] angles, ref VlSiftKeypoint k);


/// Return Type: void
///f: VlSiftFilt*
///descr: vl_sift_pix*
///k: VlSiftKeypoint*
///angle: double
[DllImportAttribute("vl.dll", EntryPoint = "vl_sift_calc_keypoint_descriptor")]
public static extern void vl_sift_calc_keypoint_descriptor(IntPtr f, IntPtr descr, ref VlSiftKeypoint k, double angle);


/// Return Type: void
///f: VlSiftFilt*
///image: vl_sift_pix*
///descr: vl_sift_pix*
///widht: int
///height: int
///x: double
///y: double
///s: double
///angle0: double
[DllImportAttribute("vl.dll", EntryPoint = "vl_sift_calc_raw_descriptor")]
public static extern void vl_sift_calc_raw_descriptor(IntPtr f, IntPtr image, IntPtr descr, int widht, int height, double x, double y, double s, double angle0);


/// Return Type: void
///f: VlSiftFilt*
///k: VlSiftKeypoint*
///x: double
///y: double
///sigma: double
[DllImportAttribute("vl.dll", EntryPoint = "vl_sift_keypoint_init")]
public static extern void vl_sift_keypoint_init(IntPtr f, ref VlSiftKeypoint k, double x, double y, double sigma);
}
}
SiftDetector类的实现代码如下所示：
using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Drawing;
using System.Runtime.InteropServices;
using Emgu.CV;
using Emgu.CV.Structure;

namespace ImageProcessLearn
{
/// <summary>
/// SIFT检测器
/// </summary>
public class SiftDetector : IDisposable
{
//成员变量
private IntPtr ptrSiftFilt;

//属性
/// <summary>
/// SiftFilt指针
/// </summary>
public IntPtr PtrSiftFilt
{
get
{
return ptrSiftFilt;
}
}

/// <summary>
/// 获取SIFT检测器中的SiftFilt
/// </summary>
public VlSiftFilt SiftFilt
{
get
{
return (VlSiftFilt)Marshal.PtrToStructure(ptrSiftFilt, typeof(VlSiftFilt));
}
}

/// <summary>
/// 构造函数
/// </summary>
/// <param name="width">图像的宽度</param>
/// <param name="height">图像的高度</param>
/// <param name="noctaves">阶数</param>
/// <param name="nlevels">每一阶的层数</param>
/// <param name="o_min">最小阶的索引</param>
public SiftDetector(int width, int height, int noctaves, int nlevels, int o_min)
{
ptrSiftFilt = VlFeatInvoke.vl_sift_new(width, height, noctaves, nlevels, o_min);
}
public SiftDetector(int width, int height)
: this(width, height, 4, 2, 0)
{ }
public SiftDetector(Size size, int noctaves, int nlevels, int o_min)
: this(size.Width, size.Height, noctaves, nlevels, o_min)
{ }
public SiftDetector(Size size)
: this(size.Width, size.Height, 4, 2, 0)
{ }

/// <summary>
/// 进行SIFT检测，并返回检测的结果
/// </summary>
/// <param name="im">单通道浮点型图像数据，图像数据不必归一化到区间[0,1]</param>
/// <param name="resultType">SIFT检测的结果类型</param>
/// <returns>返回SIFT检测结果——SIFT特征列表；如果检测失败，返回null。</returns>
unsafe public List<SiftFeature> Process(IntPtr im, SiftDetectorResultType resultType)
{
//定义变量
List<SiftFeature> features = null;      //检测结果：SIFT特征列表
VlSiftFilt siftFilt;                    //
VlSiftKeypoint* pKeyPoints;             //指向关键点的指针
VlSiftKeypoint keyPoint;                //关键点
SiftKeyPointOrientation[] orientations; //关键点对应的方向及描述
double[] angles = new double[4];        //关键点对应的方向（角度）
int angleCount;                         //某个关键点的方向数目
double angle;                           //方向
float[] descriptors;                    //关键点某个方向的描述
IntPtr ptrDescriptors = Marshal.AllocHGlobal(128 * sizeof(float));  //指向描述的缓冲区指针
//依次遍历每一阶
if (VlFeatInvoke.vl_sift_process_first_octave(ptrSiftFilt, im) != VlFeatInvoke.VL_ERR_EOF)
{
features = new List<SiftFeature>(100);
while (true)
{
    //计算每组中的关键点
    VlFeatInvoke.vl_sift_detect(ptrSiftFilt);
    //遍历每个点
    siftFilt = (VlSiftFilt)Marshal.PtrToStructure(ptrSiftFilt, typeof(VlSiftFilt));
    pKeyPoints = (VlSiftKeypoint*)siftFilt.keys.ToPointer();
    for (int i = 0; i < siftFilt.nkeys; i++)
    {
keyPoint = *pKeyPoints;
pKeyPoints++;
orientations = null;
if (resultType == SiftDetectorResultType.Normal || resultType == SiftDetectorResultType.Extended)
{
    //计算并遍历每个点的方向
    angleCount = VlFeatInvoke.vl_sift_calc_keypoint_orientations(ptrSiftFilt, angles, ref keyPoint);
    orientations = new SiftKeyPointOrientation[angleCount];
    for (int j = 0; j < angleCount; j++)
    {
angle = angles[j];
descriptors = null;
if (resultType == SiftDetectorResultType.Extended)
{
    //计算每个方向的描述
    VlFeatInvoke.vl_sift_calc_keypoint_descriptor(ptrSiftFilt, ptrDescriptors, ref keyPoint, angle);
    descriptors = new float[128];
    Marshal.Copy(ptrDescriptors, descriptors, 0, 128);
}
orientations[j] = new SiftKeyPointOrientation(angle, descriptors);  //保存关键点方向和描述
    }
}
features.Add(new SiftFeature(keyPoint, orientations));  //将得到的特征添加到列表中
    }
    //下一阶
    if (VlFeatInvoke.vl_sift_process_next_octave(ptrSiftFilt) == VlFeatInvoke.VL_ERR_EOF)
break;
}
}
//释放资源
Marshal.FreeHGlobal(ptrDescriptors);
//返回
return features;
}

/// <summary>
/// 进行基本的SIFT检测，并返回关键点列表
/// </summary>
/// <param name="im">单通道浮点型图像数据，图像数据不必归一化到区间[0,1]</param>
/// <returns>返回关键点列表；如果获取失败，返回null。</returns>
public List<SiftFeature> Process(IntPtr im)
{
return Process(im, SiftDetectorResultType.Basic);
}

/// <summary>
/// 进行SIFT检测，并返回检测的结果
/// </summary>
/// <param name="image">图像</param>
/// <param name="resultType">SIFT检测的结果类型</param>
/// <returns>返回SIFT检测结果——SIFT特征列表；如果检测失败，返回null。</returns>
public List<SiftFeature> Process(Image<Gray, Single> image, SiftDetectorResultType resultType)
{
if (image.Width != SiftFilt.width || image.Height != SiftFilt.height)
throw new ArgumentException("图像的尺寸和构造函数中指定的尺寸不一致。", "image");
return Process(image.MIplImage.imageData, resultType);
}

/// <summary>
/// 进行基本的SIFT检测，并返回检测的结果
/// </summary>
/// <param name="image">图像</param>
/// <returns>返回SIFT检测结果——SIFT特征列表；如果检测失败，返回null。</returns>
public List<SiftFeature> Process(Image<Gray, Single> image)
{
return Process(image, SiftDetectorResultType.Basic);
}

/// <summary>
/// 释放资源
/// </summary>
public void Dispose()
{
if (ptrSiftFilt != IntPtr.Zero)
VlFeatInvoke.vl_sift_delete(ptrSiftFilt);
}
}

/// <summary>
/// SIFT特征
/// </summary>
public struct SiftFeature
{
public VlSiftKeypoint keypoint;                            //关键点
public SiftKeyPointOrientation[] keypointOrientations;     //关键点的方向及方向对应的描述

public SiftFeature(VlSiftKeypoint keypoint)
: this(keypoint, null)
{
}

public SiftFeature(VlSiftKeypoint keypoint, SiftKeyPointOrientation[] keypointOrientations)
{
this.keypoint = keypoint;
this.keypointOrientations = keypointOrientations;
}
}

/// <summary>
/// Sift关键点的方向及描述
/// </summary>
public struct SiftKeyPointOrientation
{
public double angle;            //方向
public float[] descriptors;     //描述

public SiftKeyPointOrientation(double angle)
: this(angle, null)
{
}

public SiftKeyPointOrientation(double angle, float[] descriptors)
{
this.angle = angle;
this.descriptors = descriptors;
}
}

/// <summary>
/// SIFT检测的结果
/// </summary>
public enum SiftDetectorResultType
{
Basic,      //基本：仅包含关键点
Normal,     //正常：包含关键点、方向
Extended    //扩展：包含关键点、方向以及描述
}
}

MSER区域

OpenCv中的函数cvExtractMSER以及EmguCv中的Image<TColor,TDepth>.ExtractMSER方法实现了MSER区域的检测。由于OpenCv的文档中目前还没有cvExtractMSER这一部分，大家如果要看文档的话，可以先去看EmguCv的文档。

需要注意的是MSER区域的检测结果是区域中所有的点序列。例如检测到3个区域，其中一个区域是从(0,0)到(2,1)的矩形，那么结果点序列为：(0,0),(1,0),(2,0),(2,1),(1,1),(0,1)。

MSER区域检测的示例代码如下：

//MSER（区域）特征检测
private string MserFeatureDetect()
{
//获取参数
MCvMSERParams mserParam = new MCvMSERParams();
mserParam.delta = int.Parse(txtMserDelta.Text);
mserParam.maxArea = int.Parse(txtMserMaxArea.Text);
mserParam.minArea = int.Parse(txtMserMinArea.Text);
mserParam.maxVariation = float.Parse(txtMserMaxVariation.Text);
mserParam.minDiversity = float.Parse(txtMserMinDiversity.Text);
mserParam.maxEvolution = int.Parse(txtMserMaxEvolution.Text);
mserParam.areaThreshold = double.Parse(txtMserAreaThreshold.Text);
mserParam.minMargin = double.Parse(txtMserMinMargin.Text);
mserParam.edgeBlurSize = int.Parse(txtMserEdgeBlurSize.Text);
bool showDetail = cbMserShowDetail.Checked;
//计算
Stopwatch sw = new Stopwatch();
sw.Start();
MemStorage storage = new MemStorage();
Seq<Point>[] regions = imageSource.ExtractMSER(null, ref mserParam, storage);
sw.Stop();
//显示
Image<Bgr, Byte> imageResult = imageSourceGrayscale.Convert<Bgr, Byte>();
StringBuilder sbResult = new StringBuilder();
int idx = 0;
foreach (Seq<Point> region in regions)
{
imageResult.DrawPolyline(region.ToArray(), true, new Bgr(255d, 0d, 0d), 2);
if (showDetail)
{
    sbResult.AppendFormat("第{0}区域，包含{1}个顶点（", idx, region.Total);
    foreach (Point pt in region)
sbResult.AppendFormat("{0},", pt);
    sbResult.Append("）\r\n");
}
idx++;
}
pbResult.Image = imageResult.Bitmap;
//释放资源
imageResult.Dispose();
storage.Dispose();
//返回
return string.Format("·MSER区域，用时{0:F05}毫秒，参数（delta：{1}，maxArea：{2}，minArea：{3}，maxVariation：{4}，minDiversity：{5}，maxEvolution：{6}，areaThreshold：{7}，minMargin：{8}，edgeBlurSize：{9}），检测到{10}个区域\r\n{11}",
sw.Elapsed.TotalMilliseconds, mserParam.delta, mserParam.maxArea, mserParam.minArea, mserParam.maxVariation, mserParam.minDiversity,
mserParam.maxEvolution, mserParam.areaThreshold, mserParam.minMargin, mserParam.edgeBlurSize, regions.Length, showDetail ? sbResult.ToString() : "");
}

各种特征检测方法性能对比

上面介绍了这么多的特征检测方法，那么它们的性能到底如何呢？因为它们的参数设置对处理时间及结果的影响很大，我们在这里基本都使用默认参数处理同一幅图像。在我机器上的处理结果见下表：

特征用时（毫秒）特征数目

Sobel算子5.99420n/a

拉普拉斯算子3.13440n/a

Canny算子3.41160n/a

霍夫线变换13.7079010

霍夫圆变换78.077200

Harris角点9.41750n/a

ShiTomasi角点16.9839018

亚像素级角点3.6336018

SURF角点266.27000151

Star关键点14.8280056

FAST角点31.29670159

SIFT角点287.5231054

MSER区域40.629702

（图片尺寸：583x301，处理器：AMD ATHLON IIx2 240，内存：DDR3 4G，显卡：GeForce 9500GT，操作系统：Windows 7）

感谢您耐心看完本文，希望对您有所帮助。

下一篇文章我们将一起看看如何来跟踪本文讲到的特征点（角点）。

另外，如果需要本文的源代码，请点击这里下载。