In order to obtain the robust local feature descriptor, a local feature descriptor based on nonsubsampled contourlet transform (NSCT) domain was proposed, which does not need to estimate the main direction in the construction process. Firstly, the affine invariant region was detected using Hessian-Affine detector and then was normalized. Secondly, the normalized region was decomposed by NSCT to get multiple support regions of different scales and resolutions. Thirdly, the support regions were divided based on intensity orders. Finally, each sub-region was described as a vector and the local feature descriptor was obtained by aggregating the vector of each sub-region. The experimental results showed the good effectiveness of the proposed descriptor for the image with blurring, viewpoint changes, affine changes, linear illumination changes and JPEG compression changes.
