I think it's important to note that the original question was asked in 2008, and OpenCV 2.0 was released in 2009. The version 2.0 release introduced a C++ wrapper which is significantly easier to work with than the older C interface that the OP was confronted with. For my .NET project, I'm leaving all the graphic manipulation in native C++.
Try this: create a C++/CLR DLL project which links to the OpenCV libraries. The OpenCV manual describes how to do this for a Windows C++ EXE, the same steps also work for a C++/CLR DLL. Then of course the DLL exports methods which are callable from a .NET EXE.
To test it, you should be able to incorporate any of the OpenCV samples into your DLL with a little tweaking. (Add the .CPP file to your project, convert the main() function to a class member, etc. - you know the drill...) A good test candidate might be the "mat_mask_operations" sample.