Concatenate Mat in OpenCV

Question

I have a couple of images in Mat objects all with same dimensions I'd like to create one bix cv::Mat object to hold them all

So the dimension of the new matrix is: widthNew = widthOld x number of matrices, height remains unchanged.

I found that such a copy could be done using:

void cvCopy(const CvArr* src, CvArr* dst, const CvArr* mask=NULL)

but then, how could the mask be defined three different times for the three matrices?.

Regards, Moataz

Montenegro

Be careful that the max size for an IplImage is somewhere around 4000*4000. You might also have to reduce resolution of your input image – Berberine

Mask is optional, you do not need to define one. – Backstop

alright. I'll put that in mind, thnx – Montenegro

Answer 1

You use an roi to define an image which is actually a region of the destination image and then copy to that. see Copy an cv::Mat inside a ROI of another one

Allanadale

Answer 2

I think there is an easy way to do this. OpenCV has a not documented methods called hconcat() and vconcat(). The first one is for horizontal concatenation and the second one for vertical concatenation.

You can use them in this way:

Mat A, B;
... //In this part you initialize the Mat A and Mat B.

Mat H, V; //These are the destination matrices
hconcat(A, B, H);
vconcat(A, B, V);

I hope this can help.

Because

The destination can be the same as one of the sources so vconcat(out, out1, out); vconcat(out, out2, out); worked well for me in a multithreaded context (for combining output from individual threads). – Arson

Answer 3

You probably want to take look at the source code of cvShowManyImages(), which is a function that takes several images as input and concatenates them into a single 3-channel image to be displayed:

The method used is to set the ROIs of a Single Big image and then resizing and copying the input images on to the Single Big Image.

If you create a destination image big enough to hold your other images, you won't need to resize them.

If you want to know how to convert between IplImage <-> cv::Mat, check this thread.

Think

sounds interesting, thnx. I think I'll copy the idea to implement something similar using pure cv::Mat instead of IpImage – Montenegro