绝对可以使用cv :: Mat指针，但是(与所有指针一样)您必须格外小心地删除创建的对象，以再次释放分配的内存。

在您的示例代码中，尚不清楚是否将在任何地方删除其指针在函数中返回的对象。 如果不是这种情况，则cv :: Mat像素数据的引用计数将无法按预期方式工作，此外，cv :: Mat对象本身也会泄漏一些(较小的)内存。 如此简短：对于每个new cv::Mat必须有一个delete ，因此，如果外部函数中有一个new ，也许您必须为其调用一个delete (如果该函数不是某些智能类功能本身所做的，后来)。 更具体地讲， Mat* m = new Mat ( *MatPointerReturningFunc(..))将创建一个新的cv :: Mat对象，因为可以稍后将其删除，但该对象返回的指针丢失了。 尝试cv::Mat * returnedPtr = MatPointerReturningFunc(..); 跟cv::Mat * m = new cv::Mat(*returnedPtr); 如果您要创建一个新的Mat对象(或仅使用returnPtr而不创建新对象)。

这是一些示例代码，带有2个不同的Mat Ptr返回函数，以及一些有或没有泄漏访问它们的方法。

值得一提的是，使用指针的“最佳”方法通常是智能指针，由OpenCV明确给出。 但是更好的是不要对cv :: Mat对象使用Pointers，因为它不那么容易出错，并且没有太多开销。 引用计数非常好，您通常不使用int或std::vector或std::string类的指针作为指针;)

cv::Mat* MatPointerReturningFunc()

{

// this function allocates new memory by first creating a new cv::Mat object on heap (just some bytes for the Mat header)

// and second by allocating new reference counted data in that cv::Mat object (pixel memory = 640*480*3 bytes)

cv::Mat * functionMat = new cv::Mat(480, 640, CV_8UC3, cv::Scalar(rand()%255, rand()%255, rand()%255));

return functionMat;

}

cv::Mat* MatPointerReturningFunc2(cv::Mat * input)

{

// this function does not allocate new memory on the heap, just modifies the input data

cv::circle(*input, cv::Point(320, 240), 100, cv::Scalar(rand() % 255, rand() % 255, rand() % 255), -1);

return input;

}

int main(int argc, char* argv[])

{

while (cv::waitKey(0) != 'q')

{

cv::Mat * mainMat;

cv::Mat * anotherMat;

// this will not leak, because the memory which was allocated in the function, is destroyed afterwards

// and the reference counter of the openCV Mat pixel-data is decremented

if (true)

{

mainMat = MatPointerReturningFunc();

cv::imshow("main", *mainMat);

delete mainMat;

}

// this will leak because the returned Mat object will not be deleted, only the here created one - that's why the reference counting is stuck at one referenced object

if (false)

{

mainMat = new cv::Mat(*MatPointerReturningFunc());

delete mainMat;

}

// this will not leak, since all references to the Mat data are destroyed (the returned Mat and the locally created new Mat)

if (true)

{

mainMat = MatPointerReturningFunc();

cv::Mat * anotherMat = new cv::Mat(*mainMat);

cv::imshow("main", *mainMat);

cv::imshow("another", *anotherMat);

cv::imshow("another", *anotherMat);

delete mainMat;

delete anotherMat;

}

// this will not leak because all references are destroyed

if (true)

{

mainMat = MatPointerReturningFunc();

cv::imshow("main", *mainMat);

anotherMat = new cv::Mat(mainMat->clone());

cv::imshow("another", *anotherMat);

delete mainMat;

delete anotherMat;

}

// now here we work with the second function which does not allocate new memory but only modifies the input matrix

if (true)

{

// first we have to allocate memory, otherwise the input mat can't be modified ;)

mainMat = new cv::Mat(480, 640, CV_8UC3, cv::Scalar(255, 255, 255));

cv::imshow("main", *mainMat);

// this does not leak because the second functions doesn't allocate new memory:

// however we have to delete mainMat later, because we created it before.

if (true)

{

anotherMat = MatPointerReturningFunc2(mainMat);

cv::imshow("another", *anotherMat);

}

// this does not leak because we delete the new created Mat object here and the before created mainMat later

if (true)

{

anotherMat = new cv::Mat(*MatPointerReturningFunc2(mainMat));

cv::imshow("another", *anotherMat);

delete anotherMat;

}

// this does not leak because we delete the new created Mat object here and the before created mainMat later

// in fact, this way will allocate new pixel data because of the clone(), but we will free the memory with `delete anotherMat`directly

if (true)

{

anotherMat = new cv::Mat(MatPointerReturningFunc2(mainMat)->clone());

cv::imshow("another", *anotherMat);

delete anotherMat;

}

// since we created the object and allocated memory, we have to give it free here again

delete mainMat;

}

// in fact, if you WANT to use Pointers with OpenCV Mat objects you should use smart pointers.

// OpenCV has its own smart pointers, but I guess you could use any other smart pointers

// create a Mat object on the heap, allocate pixel data and access all this by a smart pointer!

cv::Ptr<:mat> smartPtr = new cv::Mat(480, 640, CV_8UC3, cv::Scalar(rand()%255, rand()%255, rand()%255) );

cv::imshow("smart pointer", *smartPtr);

// no need to delete the created Mat on the smart pointer, because the smart pointer has its own reference counting

}

return 0;

}