Non-rigid Image Registration Algorithm Based On Demons Algorithms

Non-rigid image registration is a classic issue in image processing and computer vision, it is a technology to match two or more images that are taken at different times,different views or using different sensors. It is widely used in image fusion, target orientation, image matching, accurate diagnosis of disease and so on. This paper mainly focus on image non-rigid registration based on Demons algorithm.At present, many researches have done much work in the field of image registration. Most registration algorithms assume that deformation of image is rigid. However, in fact, many objects undergo non-rigid distortion so that it is necessary to use non-rigid registration algorithm to solve local or global image deformation. Existing non-rigid registration algorithm is unsatisfied when there is a large deformation or gray uniform in image. Driving internal force of Demons algorithm come from gray gradient based on diffusion model, which is invalid and easily fall into minimum in gray uniform area, which will result in poor accuracy and lower efficiency. At the same time, when image is seriously distorted, registration accuracy and speed are not ideal. The main works and innovation are as follows:(1) Demons algorithm has been proved to be effective for non-rigid image registration,however existing Demons algorithms are limited in registration image for intensity uniformity or weak textile region, which always result in low registration accuracy and efficiency. Aiming at the problem, this paper applies R-L(Riemann-Liouville) fractional differentiation to active Demons, and proposes a new image registration based on fractional differentiation active Demons. In this paper we calculate image gradient using R-L fractional differentiation instead of traditional gradient function, not only detail feature is strengthened but also image gradient of intensity uniformity and weak textile area is enhanced, thus registration accuracy and efficiency are improved. Additionally,we give the relation curve between registration accuracy and mask parameters, which can guide one to select optimal parameters.(2) Non-rigid image registration plays an important role in computer vision and medical image.However, the typical non-rigid registration algorithms for seriously distorted deformation image lead to poor registration precision and low efficiency. Aiming at the issue, we introduce a new non-rigid image registration algorithm based on Low-rank Nystr?m approximation method and spectral feature. Firstly, we extract the spectral feature of pixel, it combine spatial feature and gray feature toform global spectral feature which are invariant to distortion. Then the spectral match method is used within diffeomorphic registration framework so that the deformation field that generated by the algorithm is smooth, reversibility, differentiability, improving registration precision; Secondly, we use Nystr?m sampling methods to speed up high dimension matrix spectral decomposition by generating Low-rank approximation matrix via randomly selected rows and columns of the Laplace matrix. Finally, we put forward image registration method based on wavelet decomposition,therefore improving the accuracy and efficiency of registration. The theory analysis and experimental results show that our algorithm can improve registration precision and efficiency.