opencv 基于sift的多张图片全景图拼接
这里是基于sift来寻找特征点经行图像的匹配的原理来进行图像拼接的，具体步骤如下：
1、利用sift特征探测器来检测出两幅图片的sift特征点
2、根据上一步提取到的特征点来提取特征向量，使用SiftDescriptorExtractor对完成特征向量提取的工作，通过他对关键点周围邻域内的像素分块进行梯度运算，得到128维的特征向量
3、进行特征向量临近匹配，找到两幅图之间相互匹配的特征点
匹配之后的结果如下图：（2张图片的）
4、在第一次匹配的基础上再进行筛选，算出匹配的特征向量之间的距离，然后只去向量距离小于2倍的最小距离的特征点
这样得到的结果为：
6、判断两幅图在结果图中的左边还是右边，这里是根据特征点在图像的位置来判断的，求出图像上特征点的x坐标大于图像宽的一半的特征点数量与总的特征点数量的比值，比值大的就是在结果图的右边
部分拼接结果：
源代码为：(为了方便测试我把图像都命名为1.jpg 2.jpg .......)
#include
#include
#include
#include
#include
#include
#include
#include
using namespace std;
using namespace cv;
Mat Stitched(Mat img1, Mat img2) {
Mat g1(img1, Rect(0, 0, img1.cols, img1.rows));
Mat g2(img2, Rect(0, 0, img2.cols, img2.rows));
cvtColor(g1, g1, CV_BGR2GRAY);
cvtColor(g2, g2, CV_BGR2GRAY);
SiftFeatureDetector siftdet;
vectorkp1, kp2;
//SIFT sift;
SiftDescriptorExtractor extractor;
Mat descriptor1, descriptor2;
FlannBasedMatcher matcher;
vector matches, goodmatches;
/*进行特征点提取*/
siftdet.detect(g1, kp1);
siftdet.detect(g2, kp2);
/* 进行特征向量提取 */
extractor.compute(g1, kp1, descriptor1);
extractor.compute(g2, kp2, descriptor2);
/* 进行特征向量临近匹配 */
matcher.match(descriptor1, descriptor2, matches);
Mat firstmatches;
/*画出第一次匹配的结果*/
drawMatches(img1, kp1, img2, kp2,
matches, firstmatches, Scalar::all(-1), Scalar::all(-1),
vector(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS);
imshow("first_matches", firstmatches);
/* 下面计算向量距离的最大值与最小值 */
double max_dist = 0; double min_dist = 1000;
for (int i = 0; i < descriptor1.rows; i++) {
if (matches[i].distance > max_dist) {
max_dist = matches[i].distance;
}
if (matches[i].distance < min_dist) {
min_dist = matches[i].distance;
}
}
cout << "The max distance is: " << max_dist << endl;
cout << "The min distance is: " << min_dist << endl;
for (int i = 0; i < descriptor1.rows; i++) {
if (matches[i].distance < 2 * min_dist) {
goodmatches.push_back(matches[i]);
}
}
Mat img_matches;
/*第二次筛选后的结果*/
drawMatches(img1, kp1, img2, kp2,
goodmatches, img_matches, Scalar::all(-1), Scalar::all(-1),
vector(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS);
imshow("good_matches", img_matches);
vector keypoints1, keypoints2;
for (int i = 0; i < goodmatches.size(); i++) {
keypoints1.push_back(kp1[goodmatches[i].queryIdx].pt);
keypoints2.push_back(kp2[goodmatches[i].trainIdx].pt);
}
/*计算单应矩阵*/
Mat H = findHomography(keypoints1, keypoints2, CV_RANSAC);
Mat stitchedImage;
int mRows = img2.rows;
if (img1.rows> img2.rows)
{
mRows = img1.rows;
}
/*判断图像在左边还是在右边*/
int propimg1 = 0, propimg2 = 0;
for (int i = 0; i < goodmatches.size(); i++) {
if (kp1[goodmatches[i].queryIdx].pt.x > img1.cols / 2) {
propimg1++;
}
if (kp2[goodmatches[i].trainIdx].pt.x > img2.cols / 2) {
propimg2++;
}
}
bool flag = false;
Mat imgright;
Mat imgleft;
if ((propimg1 / (goodmatches.size() + 0.0)) > (propimg2 / (goodmatches.size() + 0.0))) {
imgleft = img1.clone();
flag = true;
}
else {
imgleft = img2.clone();
flag = false;
}
if (flag) {
imgright = img2.clone();
flag = false;
}
else {
imgright = img1.clone();
}
/*把上边求得的右边的图像经过矩阵H转换到stitchedImage中对应的位置*/
warpPerspective(imgright, stitchedImage, H, Size(img2.cols + img1.cols, mRows));
/*把左边的图像放进来*/
Mat half(stitchedImage, Rect(0, 0, imgleft.cols, imgleft.rows));
imgleft.copyTo(half);
return stitchedImage;
}
int main() {
Mat img1 = imread("1.jpg");
Mat stitchedImage;
int n;
cout << "Dataset2" << endl;
cout << "请输入想拼接的图片数量（大于1小于18）" << endl;
cin >> n;
cout << "输入成功，开始计时" << endl;
clock_t start,finish;
double totaltime;
start=clock();
resize(img1, img1, Size(img1.cols / 4, img1.rows / 4));
for (int k = 2; k <= n; k++) {
stringstream stream;
string str;
stream << k;
stream >> str;
string filename = str + ".jpg";
cout << "正在拼接......." << filename << endl;
Mat img = imread(filename);
resize(img, img, Size(img.cols / 4, img.rows / 4));
stitchedImage = Stitched(img1, img);
img1 = stitchedImage;
}
finish = clock();
totaltime = (double)(finish - start) / CLOCKS_PER_SEC;
cout << "拼接成功" << endl;
cout << "拼接花费总时间为：" << totaltime << "秒！" << endl;
imshow("ResultImage", stitchedImage);
imwrite("ResultImage.jpg", stitchedImage);
waitKey(0);
return 0;
}