I'm working on some image manipulation code in c# and need to do some matrix operations (specifically 2D convolution). I have the code written in matlab which uses the conv2 function ... is there a library for C# / .NET that does good high-speed matrix manipulations? I'd be fine if it requires some specific GPU and does the matrix math on-GPU if that's what it takes.
Emgu is a nice C# port of OpenCV. I'm not sure about your convolution speicifically, but from this link it appears likely.
Another idea would be to create a dll with your routines that you link to with C#. I think Matlab has export/compile to C file functionality..
http://www.mathworks.com/products/netbuilder/
For GPU based work have a look at CUDA :
http://www.nvidia.com/object/what_is_cuda_new.html
.Net with CUDA :
http://www.hoopoe-cloud.com/Solutions/CUDA.NET/Default.aspx
Related
I'm writing to ask a question regarding the implementation of stationary wavelet transform in c sharp. first of all, I tried to use the swt function in Matlab by using "ReverseBior" wavelet and it works, but now I want to use C# for some reasons.
I found the swt library on Github1
Unfortunately, I'm a beginner in C# coding and I don't have any ideas about the "ReverseBior" wavelet library. therefore, I would be grateful if someone elaborates on how to find the "ReverseBior" library in C#.
As mentioned in the EMGU documentations ".Net wrapper to the OpenCV image processing library"
We depended on the functions we could find online but we really didn't figure out how to use those exact OPENCV C++ functions in the opencv documentations like blob?
How to use them in the EMGU platform?
I would suggest looking in the EmguCV source code for the OpenCV functions you are looking for and see how they are wrapped. It is all there.
There is also the need to move data between the managed side and the unmanaged side.
Doug
EmguCV provides a class called CvInvoke which wraps a lot of the basic opencv functionality. The previous link goes for the EmguCV 3.4.3 documentation. Maybe you need another version. Anyway I recommend to also check other classes aside from CvInvoke. The API documentation should be a good starting point.
I'm trying to make some geometric transformations (e.g. scale,shear) on an image already loaded in a bitmap object
I then found a Built-in Matrix class in C# , I used it but I'm not sure on how to initialize a 2x2 matrix or a column vector etc.
The 4th overloaded constructor forces me to enter 6 values representing 3x2 ,but that I don't want!
I used this reference but couldn't find an answer
https://msdn.microsoft.com/en-us/library/system.drawing.drawing2d.matrix(v=vs.110).aspx
As #Daniel pointed out, System.Drawing.Drawing2D.Matrix is for visual geometrical transformations only, not mathematical.
If you'd like mathematical transformations I'd recommend you take a look at Math.NET or what I'd recommend, OpenCV library. OpenCV is written in C++, but has an excellent C# wrappers. It can be easily installed via Nuget.
I'm currently searching for a C# image recognition library.
What I want to do:
I want to write a function that scans an image and returns if another image is part of it. Or at least something that looks familiar in case that the angles of the two objects are different.
The link to a possible library and a short code example would be great!
Thank you in advance!
Since you didn't mention that you are only looking for free libraries, here are some paid ones:
MVTech HALCON
Cognex VisionPro
Both have demo versions and quite good .Net wrappers bundled to the SDK, and I think both have the functionality you need. In Halcon, you might want to try different matching algorightms (gray value based, descriptor based, etc.), while in VisionPro PatMax or PatQuick might suit your needs. But obviously you have to try which one is the best for your specific problem.
EmguCV (http://www.emgu.com/wiki/index.php/Main_Page) is a good .NET OpenCV wrapper. It has a bunch of sample projects bundled. Run samples and you will get the idea of what can be done and how.
The Accord.NET library is not actually an image recognition tool set, however it provides the base for what you are aiming for. It contains many Imaging classes required for building an image recognition system. Accord.NET is LGPL licensed, except for some parts of it (e.g. its FFmpeg wrapper project).
in openCV I regularly use cv::Mat for almost everything. Now, I need to use emgu CV and use the Matrix-object in stat, but some functions are not supported?!... may I use the image-class instead?
When to use image and when to use matrix in emgu CV?
P.S.: Currently I'm looking for a way to define a ROI on a matrix but didn't find a way without copying the data.
Version: Emgu.CV-2.4.2
As of OpenCV 3.0, OpenCV is phasing out IplImage. EmguCV's Image<,> is a wrapper around IplImage, so Image<,> is being phased out of EmguCV.
Use Mat and CvInvoke wherever possible. Avoid Image<,> and its methods.
I know that in the latest version of OpenCV, the Mat object is prefered to iplImage for a lot of reasons.
In EmguCV, things are different. I think that the rule of thumbs is to use Image<> when the data itself is supposed to be an image and Mat<> when you need to work with a matrix. Work with what you need and the emguCV object should have the right contructor/function for your needs.
For the ROI, the comment by sumeet is very good.
Hope it helps!