earticle

영어

Image registration refers to finding a geometrical transformation that correspond any point from one image to its homologous on the other image. There are several similarity measures that are classified in two groups based on features and intensity. In medical imaging, accuracy of registration algorithm is important. Since intensity-based methods, are more accurate than feature based ones, we select intensity-based registration; But intensity based methods usually need to global or local similarity measure optimization. Due to large search space, global methods optimization is time-consuming and when image irregularities are large, local methods cannot reach to an optimum amount. Despite these challenges, we found that learning based methods can be an appropriate policy to overcome these problems.
Accordingly, in this paper, instead of using a fixed similarity measure, learning based similarity measure methods will present. Using the presented approaches in this paper can have been an effective role in analyzing and evaluating multi modal medical image registration and will increase three main functional measures – accuracy, speed and robustness – in medical image registration.