Résumé : Real-time cardiac MRI appears as a promising technique to evaluate the mechanical function of the heart. However, ultra-fast MRI acquisitions come with an important signal-to-noise ratio (SNR) penalty, which drastically reduces the image quality. Hence, a real-time denoising approach would be desirable for SNR amelioration. In the clinical context of cardiac dysfunction assessment, long acquisitions are required and for most patients the acquisition takes place with free breathing. Hence, it is necessary to compensate respiratory motion in real-time. In this article, a real-time and interactive method for sequential registration and denoising of real-time MR cardiac images is presented. The method has been experimented on 60 fast MRI acquisitions in five healthy volunteers and five patients. These experiments assessed the feasibility of the method in a real-time context.