Research Of The Texture Classification Algorithm

Texture classification as an important aspect in texture analysis has brought great influence in numerous applications. People have proposed lots of feature extraction and classification algorithms and achieved great successful in the last few years, but there are also many disadvantages that restrict the applications to some extent. Rotation invariant and antinoise texture classification is a topic to be researched deeply. In the foundation of the deeper research on the existing algorithms, we mainly research on the following aspects in this paper:(1)Texture classification based on Gabor filters and the dual-tree complex wavelet transform (DT-CWT). This section is for testing the performance of the dual-tree complex wavelet transform (DT-CWT) and support vector machines (SVM). Compute the energies or local energy functions of the subbands in different scales and orientations acquired by filtering the texture images with Gabor transform or DT-CWT to build feature vectors. The SVM is mainly used to classify while the classifiers of k-means, BP neural network and LVQ are also used in the experiments for comparison.(2)Rotation invariant texture classification based on rotation-translation. Rotation invariant texture classification based on rotation-translation is carried out by two ways in this paper. First, the rotated texture image is transformed by log-polar or Radon transform to convert the rotation to translation, the rotation invariant feature vector is acquired by offsetting the shift of the translated texture image using discrete stationary wavelet transform (SWT) or DT-CWT that are shift invariant. Second, compute the energies of the subbands acquired by transforming the texture image using Curvelet transform to convert the rotation to the translation of energies in the same scale, then extract the rotation invariant feature vectors of isotropic, anisotropic and circular shift. Classify thefeature vectors using SVM and compare the three methods in the first aspect.(3)Texture classification with antinoise property based on the LBP features of Curvelet dominant energy subbands. Aiming at the disadvantages of the exis- ting algorithms that don't have performance of antinoise and can't description textures with bigger textons effectively, a rotation invariant texture classification algorithm based on the LBP features of dominant Curvelet energy subbands is proposed. The images are classified by SVM at last. The rotation invariant fea- ture vectors acquired above are multi-resolutions and have stronger antinoise property, the LBP operators of the same size can character the original texture in lager region.