Geometric models of point-based and region-based features are potent tools for cardiac image registration. This paper presents a new region-based approach which uses Gaussian process to specify the breathing and cardiac motion in magnetic resonance imaging. To take the advantage of the fact that each kind of motion has its specific spatial influence we interpret the motions with distinct Gaussian processes. By elimination of respiratory move in a sequence of frames we can determine positions of regions which express edges of cardiac motion. From the data the Gaussian distribution can be applied again for a statistical model of heart regions which were observed in movement between frames. The intensity divergence corresponding to the region of cardiac motion depicts the structure of the kinetics of heart. In particular, validation study of the method was involved with a database of cardiac sequence of images which cover both heart and respiratory motions. The study has indicated that the Gaussian processes do grant respiratory move to be subtracted, a step towards effective registration of cardiac motions.