Study Of Rotating Invariant Texture Classification Algorithm

With the widely use of texture analysis, texture classification is becoming an active research top. But most traditional algorithm has the drawback of low classification rate and complex computation. This is asking for an improve. This paper do research in the rotation invariant texture classification of two algorithm, based on LBP and filter banks, and make a comparison on them.(1)Based on the drawback of the joint histogram LBP and VAR, we introduce a new operator called LBPV, which even though is a simplified operator of joint distribution but can efficient address the problems of it.(2)In order to avoid the misclassification caused by useing the LBP operator, we research a new matching method called global matching, which is a hybrid method with both rotation invariant and varint LBP pattern, it can improve the classificatin rate effectively.(3)The global matching uses exhaustive search which makes the computational cost high, so we estimate the principal orientations first, then match along the principal orientations only. We use the LBP pattern frequency to estimate principal orientations. The results of experiment prove that this approach can reduce the computation effectively.(4)We use a new method, based on rotation invariant LBP, to reduce the feature dimensions to reduce the computation.(5)The algorithm based on filter banks(MR8) that we choose is similar to LBP method, in this paper we compare the two algorithms and make a summary.