Get frame video with opencv

Question

This is my code to get frame from video. I want to show 2 frame in video with looping and condition.

    int main( int argc, char** argv )
    {    

    string fileName = "E:\\Tugas Akhir\\Video Master\\city_1.avi";
    Mat image1;
    Mat image2;
    Mat frame;
    cv::Mat result;
    VideoCapture cap(fileName); 
    if(!cap.isOpened())  
        return -1;

    Mat edges;
    for(int loop=0;;loop++)
    {
        //std::cout<<loop<<endl;
        cap >> frame; // get a new frame from camera
        if(loop>0 && (loop%20)==0){             
            //std::cout<<"frame 2"<<endl;
            image2=frame;                               
            **imshow("image2",image2);**
            break;      
        }else if(loop==0){          
            image1=frame;
            **imshow("image1",image1);**
            //std::cout<<"frame 1"<<endl;

        }
        //loop++;


        if(waitKey(30) >= 0) break;
    }


 waitKey(0);
 return 0;
 }

And here's the result, 2 windows with 2 different image

but when i change imshow("image1",image1) method position ...

           if(loop>0 && (loop%20)==0){              
            //std::cout<<"frame 2"<<endl;
            image2=frame;

            **imshow("image1",image1);**
            **imshow("image2",image2);**
            break;      
        }else if(loop==0){          
            image1=frame;

            //std::cout<<"frame 1"<<endl;

        }

image1 windows show same picture with image2,

i don't know why it show strange result, please tell me how to fix it, thank you

Answer 1

an assignment like this:

image1 = frame;

does only a shallow copy. the Mat struct gets copied, the pixels are shared

so, in your 2nd example you're overwriting image1 with the current frame. if you want to 'cache' Mats, use:

image1 = frame.clone(); // deep copy

Answer 2

What you're facing is essentially the lack of support for copy-on-write in OpenCV Mat's overloaded = operator. It basically means that image1 and frame will share their data.

A convenient way of seeing this is, when you write image1 = frame , called "shallow copying", you're creating a reference image1 to the Mat frame . So after 20 iterations when you display image1 , you're actually displaying frame itself.

This wouldn't be the case if you'd written something like image1 = frame.clone() , because in that case you're actually making a separate copy of frame .

Also check this out.