Anisotropic Optical Flow And Its Application To Brain MR Image Registration

This paper analyses the influences of noise and bias field in brain MR image registration, and researches the optical flow model and PDE’s based image denoising. Then purposes an improved anisotropic image registration model based on the optical flow. This paper focus on the brain MR image registration after discuss the development history and the technology shortages of the PDE’s based image denoising and optical flow model, and for its research and improvement, this article works on1, Due to the use of isotropic regularization in the original optical flow will cause the blur of image, so the results of image registration will lose image structure when we replace the displacement field with original optical flow. This paper introduces the anisotropic regularization to the original optical flow model will keep the structural information in the image registration.2, The noise in the MR image will affect the result of image process, this article use the nonlocal mean to decrease the effects of noise in the calculation of optical flow, in order to increase the accuracy of registration.3, The intensity inhomogeneity of MR image doesn’t comply the basic conjecture of optical flow, which will get a wrong result. This paper coupled the optical flow model with bias field model, The bias field will be calculated during registration. And the introduction of the bias field recovery model will reduce the effects of intensity inhomogeneity. The implementation shows the improved model can obtain better registration results.