Rotation Invariant Texture Classification Using Principal Direction Estimation

摘要

The rotation invariant texture classification is an important application of texture analysis. A rotated texture is often perceived by the changed dominant direction. This paper proposes an effective rotation-invariant texture classification method by combining the local patch based method with the orientation estimation. For a texture sample, the Principal component analysis is applied to its local patch to estimate the local orientation, and then the dominant orientation is determined with the maximum value of the local orientation distribution. In order to extract the feature vector, each local patch is rotated along the dominant orientation after circular interpolation. By using the random projection, the local gray value vector of a patch is mapped into a low-dimensional feature vector that is placed in the bag of words model, together with local orientation feature. The simulation experiments demonstrate the proposed method has a comparable performance with the existing methods.