I did not find the software to finish this job. From the author’s thesis, the software may be built by themselves. The software should have the function which can detect the key points and generate the feature descriptors.
I don’t know the method without the “training” images. But I am very interested in that.