A pilled nonwoven fabric image consists of brightness variations caused by high frequency noise, randomly distributed fibers, fuzz and pills, fabric surface unevenness, and background illumination variance. They have different frequency and space distributions and thus can be separated by the two-dimensional dual-tree complex wavelet transform reconstructed detail and approximation images. The energies of the six direction detail sub-images, which capture brightness variation caused by fuzz and pills of different sizes, quantitatively characterize the pilling volume distribution at different directions and scales. They are used as pilling features and inputs of neural network supervised classifier. The initial results based on a nonwoven wool fabric standard pilling test image set, the Woolmark$^{(R)}$, SM 50 Blanket set, suggest that this objective pilling evaluation method developed by the combination of pilling identification, characterization method and neural network supervised classifier is feasible.