Texture classification

which is the key technology for computer vision and equipment monitoring

plays an important role in industrial production.Texture classification improves not only the efficiency of production but also product quality and reliability. For signal processing technology

a new method with a higher efficiency is provided for image texture classification. The projection of texture as a time domain signal is analyzed using empirical mode decomposition. The main projection direction is selected

and then two-dimensional signals are converted into one-dimensional signals by Radon transform. The end effects of the projection signal are restrained and divided into intrinsic mode functions (IMFs). After the statistical characteristics of the IMFs are calculated

they are compressed and simplified using principal component analysis to reduce their dimension. Once the effect of the principal characteristics is accessed with a support vector machine

classification is realized. Texture classification experiments are conducted in multiple directions and at multiple scales using Brodatz and KTH-TIPS datasets. A statemonitoring system for printing machines based on texture has been established. The calculation velocities of the proposed Radon-EMD

GLCM

and Gabor were compared using several Brodatz images

and the average times of the three texture analysis methods are approximately 5 s

9.5 s

and 24 s

respectively. The study proved that IMFs of projection are good at texture classification and have the significant advantage of computational simplicity. This method obtains a good classification result in multiple directions and at multiple scales. Thus

the method has good computational efficiency.