AForge.NET, will strongly recommend it. Step-by-step tutorials? They can be helpful, but how badly do you need them? If you really cannot leave without them, you probably should not go in for such a difficult topic as computer vision. I consider this factor negligible. More important some theoretical skills, knowledge of at least near-elementary mathematics, at least some related fields. Also, it will need some "recognition feeling", intuition, which can be developed with some experience, not so much of trial-and-error kind of experience (unavoidable anyway), but more of thinking experience.
As to regular documentation of AForge.NET, it's good enough.
Also, you can consider Open CV; you can find .NET wrappers, such as OpenCVDotNet, OpenCV.NET or EmguCV.
http://en.wikipedia.org/wiki/OpenCV[
^],
http://opencv.org[
^],
https://code.google.com/p/opencvdotnet[
^],
https://bitbucket.org/horizongir/opencv.net[
^],
http://www.emgu.com/wiki/index.php/Main_Page[
^] (probably there are more).
I tried to use some big-name recognition packages, very expensive, and what? Their licensing schemes killed me. It's not just money, this is like wearing handcuffs, not normal working convenience. A lot of hype, and still quality problems. A typical problem of the professionals in the field is strong focus on applied mathematics at the expense of programming culture, which translates into quality problems. Keeping this in mind, open-source products have unmatched benefits, even with generous budget…
—SA