An improved non-rigid image registration approach

Abstract

With the rapid development of image registration technology, it is being widely used in the fields of medical image processing, remote sensing image analysis, computer vision, and others. Image registration involves two or more images that contain the same object that are obtained under different conditions. Geometric mapping between images is realized by spatial geometric transformation, so that the points in one image can be related to their corresponding points in the other. Compared with rigid transformations, non-rigid transformations usually have severe local distortions and obvious nonlinear characteristics. So, it is difficult to describe non-rigid transformations using a unified transformation model. For this reason, non-rigid image registration has always been an issue and a source of difficulty in the field of computer vision. To solve this problem, an improved optical-flow-model algorithm was proposed to more accurately estimate the optical flow field. First, the original variational optical flow model was improved. To prevent blurring and preserve the edge and detail features of images, a new anisotropic regular term was proposed to replace the original homologous diffusion term. Then, to remove optical flow outliers, a non-local smoothness term was introduced that contained neighborhood information. Moreover, a weight function that combines image-structure and optical-flow information was added to reduce the loss of detail caused by over-smoothing and to improve robustness. Finally, to solve the displacement field and realize the automatic registration of non-rigid images, an alternating minimization method and pyramid hierarchical iteration strategy were utilized. To verify the effectiveness of the proposed algorithm, subjective and objective evaluation values such as the peak signal-to-noise ratio (PSNR) and normalized mutual information (NMI) were adopted to analyze the registration results. Compared with state-of-the-art methods, experimental results reveal the robustness and ideal registration effects of the proposed method on different types of non-rigid images.

Keywords

• non-rigid image registration
• optical flow model
• anisotropic regularization term
• non-local smoothness term
• alternate minimization