OpenCV copyTo、clone、“=”与拷贝构造函数的区别

opencv中为矩阵复制提供了copyTo函数、clone函数、重载运算符和拷贝构造函数，用法非常简单：

 Mat srcimage = imread("1.jpg"); Mat firstimage,secondimage,thirdimage; srcimage.copyTo(firstimage); secondimage = srcimage.clone(); thirdimage = srcimage; Mat fourthimage(srcimage); 

但是在他们四者间存在着较大的区别，我们先写个简单的代码测试一下：

 Mat srcimage = imread("2.jpg"); Mat firstimage,secondimage,thirdimage; double t1 = (double)getTickCount(); srcimage.copyTo(firstimage); t1 = (double)getTickCount() - t1; double t2 = (double)getTickCount(); secondimage = srcimage.clone(); t2 = (double)getTickCount() - t2; double t3 = (double)getTickCount(); thirdimage = srcimage; t3 = (double)getTickCount() - t3; double t4 = (double)getTickCount(); Mat fourthimage(srcimage); t4 = (double)getTickCount() - t4; printf("copyTo execution time = %lfms\n", t1*1000. / getTickFrequency()); printf("clone execution time = %lfms\n", t2*1000. / getTickFrequency()); printf(" = execution time = %lfms\n", t3*1000. / getTickFrequency()); printf(" 构造函数 execution time = %lfms\n", t4*1000. / getTickFrequency()); getchar(); 

为了让效果明显，选择了一张很大的图片，这样图片的尺寸是3120*4160，实验结果如下：

我们可以看到，copyTo函数、clone函数占用时间差不多，但是“= ”运算符与拷贝构造函数用的时间却非常少，这是为什么呢？

造成这样的情况的原因是因为Mat的数据类型以及它的数据组成造成的。

Mat类

从上面的话我们可以看到，copyTo函数、clone函数拷贝的不仅仅是信息头，还有矩阵本身，而“= ”运算符与拷贝构造函数仅仅拷贝了信息头，他们指向的其实是一个矩阵，也就是在上的程序中，我们改变srcimage ，thirdimage和fourthimage中的任何一个，另外两个也会跟着变化，我们再改一下代码测试一下（为了屏幕能放下，我们把图片换一下）：

 Mat srcimage = imread("1.jpg"); Mat firstimage,secondimage,thirdimage; imshow("result1",srcimage); double t1 = (double)getTickCount(); srcimage.copyTo(firstimage); Point center = Point(55,55); int r = 10; circle(firstimage,center,r,Scalar(0,0,255),-1); t1 = (double)getTickCount() - t1; imshow("result2",firstimage); double t2 = (double)getTickCount(); secondimage = srcimage.clone(); t2 = (double)getTickCount() - t2; double t3 = (double)getTickCount(); thirdimage = srcimage; t3 = (double)getTickCount() - t3; double t4 = (double)getTickCount(); Mat fourthimage(srcimage); t4 = (double)getTickCount() - t4; Point a = Point (0,0); Point b = Point (thirdimage.cols,thirdimage.rows); line(thirdimage,a,b,Scalar(255,0,0)); printf("copyTo execution time = %lfms\n", t1*1000. / getTickFrequency()); printf("clone execution time = %lfms\n", t2*1000. / getTickFrequency()); printf(" = execution time = %lfms\n", t3*1000. / getTickFrequency()); printf(" 构造函数 execution time = %lfms\n", t4*1000. / getTickFrequency()); imshow("result3",srcimage); waitKey(0); getchar(); 

可以看到，在原图刚刚读入的时候，显示了窗口result1，此时显示的就是读入的图片，firstimage是由srcimage通过copyTo的方法拷贝的，我们把firstimage画一个圆，然后用result2窗口显示，thirdimage是由srcimage通过“=”运算符拷贝的，我们将thirdimage画个线，并用result3再一次显示srcimage，然后我们可以看到，result3中srcimage已经变了，变成了和thirdimage一样的效果，说明之前的说法是正确的，他们共有了同一个矩阵。