Medical Image Registration Using Low-Rank Matrix Recovery

In recent years, the medical image registration technique has developed rapidly. It’s general for patients to image time and again in one or more modes clinically, and we can observe the growth of the infection focus, the effect of the drug treatment and comparing the changes before and after operation. Medical image fusion and medical image registration techniques should be used in these processes, and the medical image registration is the basis of the image fusion.The medical image registration is mainly divided into registration based on image characteristics and registration based on image gray scale. The traditional medical image registration algorithms based on image characteristics generally embrace four parts:composing the characteristic space by extracting the image feature information, ascertaining the geometric transformation; doing interpolate operation; choosing the similarity measures; achieving the optical value of the similarity measure by taking optimizing measures. Among them, the feature extraction method with good invariance and accuracy is the key problem. The registration algorithm based on image characteristics is fast and has a small amount of calculation. Nevertheless, he registration algorithm based on image gray scale can avoid the serious consequence caused by image features extraction, accordingly its accuracy is higher.An algorithm of medical image registration, which is a registration method based on image gray scale, using low-rank matrix recovery which seeks a similarity transformation for purpose of image registration using low-rank matrix recovery methods has been proposed in this letter. Due to the MRI medical images of the same person has a certain similarity, we can do some geometric transformation to these MRI images and then use them to compose a low rank matrix, and then decompose the low rank matrix into low rank parts and sparse noise matrix part. Therefore, medical image registration issue renders into a low-rank matrix recovery problem. We realize the medical image registration as the augmented Lagrange multiplier optimization method used to solve the optimization model.