Research On The Application Of L1 Norm In Image Elastic Transformation

Image registration is an important research topic in the field of image processing, image elastic transformation based on marking is an important content in the research of image registration and is widely used in medical image analysis, remote sensing imaging, biological information recognition field. This study focuses on how to establish a performance space transformation function is superior, the robustness and topology preservation in the image registration problem, the main work includes three parts:First, innovation to L1 norm applied to elastic image transformation, the problem solving based on the elastic transformation function of radial basis function is transformed to an optimization problem, in the cost function of the optimization problem, using the L1 norm regularization to elastic constraint variable transfer function coefficients using a fast iterative shrinkage thresholding algorithm(FISTA) to solve optimization problems. Experiments on simulated images and real images show that the proposed method is robust to noise, and the registration accuracy is better.Secondly, for larger twist deformation image registration problem, in the cost function to solve the transform function of the join smoothness constraint, obtained with the smoothness of the transformation function, thereby inhibiting the elastic deformation of the distorted excessively fierce, in a certain extent obstacle free topology does not maintain the phenomenon occurred. The experimental results show that the proposed method can optimize the image distortion, and a good compromise between the registration accuracy and the distorted energy is achieved.Finally, the L1 norm regularization is introduced to robust point matching RPM algorithm, solve the point correspondence automatically determines the problem, and the RPM algorithm prone to topology does not maintain a phenomenon, introducing the smoothness constraint, to improve the accuracy of image registration. Through the experiments of different kinds of medical images demonstrate that the scheme has good registration effect.