By looking over the code and the description, it seems the filters are being crated in the frequency domain and then applied in the spatial domain. Please correct me if I am wrong but this is WRONG. There are two RIGHT approaches. The first is that one needs to take the inverse FT of the filter and then convolve it with the image. The second it that one needs to convert the image into the frequency domain, multiply the spectra and then take the inverse of the result.
This submission definitely stands out. A great and versatile package. For those interested in classification I invite you to visit http://prtools.org/index.html