33-SIFT特征匹配（EmguCV学习）

emm

懒了很久，还有最后一点儿，全部整理完~

Record

---

1、SIFI算子使用流程：

1. 初始化SIFT向量，调用Detect()方法识别两张图片各自的特征点；
2. 使用SIFT的Compute()方法计算计算两张图片各自特征点的特征描述符；
3. 使用匹配器Matcher进行特征描述符匹配（BF与Flann），以其中一幅图片为模板，在另一幅图片中寻找匹配的特征点；
4. 匹配结果保存在DMatch结构体变量中，设置距离阈值对匹配结果进行筛选，得到好的匹配点；
5. 使用DrawMatch()方法进行绘制，显示匹配结果；

---

2、BF暴力匹配器在EmguCV中通过BFMatcher类进行实现，使用Add()方法添加模板的特征描述符，使用 KnnMatch() 方法寻找计算另一幅图像中的k个匹配特征点；

---

3、可以参考：https://www.cnblogs.com/Jessica-jie/p/8622449.html

---

4、PPT:

Code

BF匹配：

using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Threading.Tasks;

using Emgu.CV;
using Emgu.CV.CvEnum;
using Emgu.CV.Structure;
using Emgu.Util;
using Emgu.CV.Util;
using System.Drawing;
using Emgu.CV.XFeatures2D;      //包含SIFT类
using Emgu.CV.Features2D;       //包含Features2DToolBox类

namespace lesson33
{
    class Program
    {
        static void Main(string[] args)
        {
            Mat srcImg1 = CvInvoke.Imread("1.jpg");
            Mat srcImg2 = CvInvoke.Imread("2.jpg");

            CvInvoke.Imshow("src1", srcImg1);
            CvInvoke.Imshow("src2", srcImg2);

            SIFT sift = new SIFT(1000);
            //计算特征点
            MKeyPoint[] keyPoints1 = sift.Detect(srcImg1);
            MKeyPoint[] keyPoints2 = sift.Detect(srcImg2);
            //绘制特征点
            Mat sift_feature1 = new Mat();
            Mat sift_feature2 = new Mat();
            VectorOfKeyPoint vkeyPoint1 = new VectorOfKeyPoint(keyPoints1);
            VectorOfKeyPoint vkeyPoint2 = new VectorOfKeyPoint(keyPoints2);
            Features2DToolbox.DrawKeypoints(srcImg1, vkeyPoint1, sift_feature1, new Bgr(0, 255, 0),
                Features2DToolbox.KeypointDrawType.Default);
            Features2DToolbox.DrawKeypoints(srcImg2, vkeyPoint2, sift_feature2, new Bgr(0, 255, 0));
            //显示绘制结果
            CvInvoke.Imshow("sift_feature1", sift_feature1);
            CvInvoke.Imshow("sift_feature2", sift_feature2);
            //计算特征描述符
            Mat descriptors1 = new Mat();
            Mat descriptors2 = new Mat();
            sift.Compute(srcImg1, vkeyPoint1, descriptors1);
            sift.Compute(srcImg2, vkeyPoint2, descriptors2);
            //使用BF匹配器进行暴力匹配
            BFMatcher bFMatcher = new BFMatcher(DistanceType.L2);
            VectorOfVectorOfDMatch matches = new VectorOfVectorOfDMatch();
            //添加特征描述符
            bFMatcher.Add(descriptors1);
            //k最邻近匹配
            bFMatcher.KnnMatch(descriptors2, matches, 2, null);
            //寻找匹配结果中距离的最值
            double min_dist = 100, max_dist = 0;
            for(int i = 0; i < descriptors1.Rows;i++)
            {
                if(matches[i][0].Distance > max_dist)
                {
                    max_dist = matches[i][0].Distance;
                }
                if(matches[i][0].Distance < min_dist)
                {
                    min_dist = matches[i][0].Distance;
                }
            }
            //对BF匹配结果进行筛选
            VectorOfVectorOfDMatch good_matches = new VectorOfVectorOfDMatch();
            for(int i = 0; i < matches.Size;i++)
            {
                //符合条件的匹配点进行存储
                if(matches[i][0].Distance < 1.5 * min_dist)
                {
                    good_matches.Push(matches[i]);
                }
            }
            //绘制匹配点
            Mat result = new Mat();
            Features2DToolbox.DrawMatches(srcImg1, vkeyPoint1, srcImg2, vkeyPoint2, good_matches,
                result, new MCvScalar(0, 255, 0), new MCvScalar(0, 0, 255));
            //显示匹配结果
            CvInvoke.Imshow("match-result", result);

            CvInvoke.WaitKey(0);
        }
    }
}

---

FLANN匹配：

using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Threading.Tasks;

using Emgu.CV;
using Emgu.CV.CvEnum;
using Emgu.CV.Structure;
using Emgu.Util;
using Emgu.CV.Util;
using System.Drawing;
using Emgu.CV.XFeatures2D;      //包含SIFT类
using Emgu.CV.Features2D;       //包含Features2DToolBox类
using Emgu.CV.Flann;            //包含快速最邻近匹配
namespace lesson33
{
    class Program
    {
        static void Main(string[] args)
        {
            Mat srcImg1 = CvInvoke.Imread("11.jpg");
            Mat srcImg2 = CvInvoke.Imread("22.jpg");

            CvInvoke.Imshow("src1", srcImg1);
            CvInvoke.Imshow("src2", srcImg2);

            SIFT sift = new SIFT(1000);
            //计算特征点
            MKeyPoint[] keyPoints1 = sift.Detect(srcImg1);
            MKeyPoint[] keyPoints2 = sift.Detect(srcImg2);
            //绘制特征点
            Mat sift_feature1 = new Mat();
            Mat sift_feature2 = new Mat();
            VectorOfKeyPoint vkeyPoint1 = new VectorOfKeyPoint(keyPoints1);
            VectorOfKeyPoint vkeyPoint2 = new VectorOfKeyPoint(keyPoints2);
            Features2DToolbox.DrawKeypoints(srcImg1, vkeyPoint1, sift_feature1, new Bgr(0, 255, 0),
                Features2DToolbox.KeypointDrawType.Default);
            Features2DToolbox.DrawKeypoints(srcImg2, vkeyPoint2, sift_feature2, new Bgr(0, 255, 0));
            //显示绘制结果
            CvInvoke.Imshow("sift_feature1", sift_feature1);
            CvInvoke.Imshow("sift_feature2", sift_feature2);
            //计算特征描述符
            Mat descriptors1 = new Mat();
            Mat descriptors2 = new Mat();
            sift.Compute(srcImg1, vkeyPoint1, descriptors1);
            sift.Compute(srcImg2, vkeyPoint2, descriptors2);
            //快速最邻近匹配Flann
            IIndexParams id = new LinearIndexParams();
            SearchParams sp = new SearchParams();
            FlannBasedMatcher flannBasedMatcher = new FlannBasedMatcher(id, sp);
            //添加模板
            flannBasedMatcher.Add(descriptors1);
            VectorOfVectorOfDMatch matches = new VectorOfVectorOfDMatch();
            flannBasedMatcher.KnnMatch(descriptors2, matches, 2, null);

            //筛选匹配结果
            Mat mask = new Mat(matches.Size, 1, DepthType.Cv8U, 1);
            mask.SetTo(new MCvScalar(255));
            Features2DToolbox.VoteForUniqueness(matches, 0.3, mask);    //去除重复匹配

            Mat result = new Mat();
            Features2DToolbox.DrawMatches(srcImg1, vkeyPoint1, srcImg2, vkeyPoint2, matches, result, new MCvScalar(0, 255, 0),
                                new MCvScalar(0, 0, 255), mask);
            CvInvoke.Imshow("match-result", result);
            CvInvoke.WaitKey(0);
        }
    }
}

效果

BF匹配：



Flann匹配：