Emgu cv high resolution images stitching Issue

Question

I am using EmguCV library to stitch images. Its working fine for small images but getting exception when processing high resolution images or images above 20MB size or even if I try to process more than 30 images it fails.

Libraries I am using

Emgu.CV.UI
Emgu.CV.UI.GL
Emgu.CV.World
opencv_core2410
opencv_imgproc2410

Code

List<Image<Bgr, Byte>> sourceImages = new List<Image<Bgr,byte>>();
foreach (string path in ImgPath)
    sourceImages.Add(new Image<Bgr, Byte>(path));
using (Stitcher stitcher = new Stitcher(false))
{
    using (VectorOfMat vm = new VectorOfMat())
    {
        Mat result = new Mat();
        vm.Push(sourceImages.ToArray());
        stitcher.Stitch(vm, result);
        if (result.Bitmap != null)
        {
            result.Bitmap.Save(Application.StartupPath + "\\imgs\\StitchedImage.png");
        }
        else
        {
            MessageBox.Show("Some thing went wrong"); return null;
        }
    }
}

Exception

((Emgu.CV.MatDataAllocator)(result))._dataHandle.Target' threw an exception of type 'System.InvalidOperationException

Image

Answer 1

I was fairly certain that you are running into a memory issue and so I went ahead and made a simple console app targeting .Net 4.7.2, using the latest EmguCV package from NuGet , which is version 3.4.3.3016, and using the code below on sample data from Adobe, which can be downloaded here . If I compile as "AnyCPU" with "prefer 32 bit" and run this code against the rio image set (I loaded up the png's for the test purposes) and let it run the memory will slowly jump up until it hits close to 3 GB and then shortly thereafter crashes giving the exception about the refcount. Pretty clearly a memory issue. I then recompiled targeting 64 bit and was able to successfully run the code. The memory usage peaked out around 6 GB. So, with that in mind, I would be fairly certain that your issue is also memory related. You have yet to answer the question on whether you are building a 64 bit app or not, but based on what you are seeing, I would guess that you are not. So, the solution to your problem is to compile as 64 bit and then be sure you have enough memory. With the rio test set it jumped up to close to 6 GB. Without having your images I can't tell you how large it might grow, but these type of operations are pretty memory intensive - so more is better. This would explain both the issue with large image files and the issue with a large number of small image files. I was able to successfully process sets of images that were between 10 and 20 images using a 32 bit build, but as soon as I moved to the 50+ image sets it would only work with a 64 bit build due to the memory requirements.

var images = Directory.EnumerateFiles(@"C:\test\adobe\rio", "*.png", SearchOption.TopDirectoryOnly).Select(x => new Mat(x)).ToArray();
using(var stitcher = new Stitcher(false))
{
    using (var vm = new VectorOfMat(images))
    {
        var result = new Mat();
        stitcher.Stitch(vm, result);
        result.Bitmap.Save(@"C:\test\adobe\rio_stitched.jpg", System.Drawing.Imaging.ImageFormat.Jpeg);
        result.Dispose();
    }
}
foreach (var image in images) { image.Dispose(); }