Frame Rate Up-Conversion in Echocardiography Images, Using Manifold-Learning and Image Registration

In this paper, we propose a new temporal frame interpolation algorithm for frame rate up-conversion (FRUC) in echocardiography images. This algorithm employs a combination of dimension reduction techniques and image registration to increase frame rate. If the distance between two successive frames of a video be great, motion jerkiness will appear between them and visual quality of the video will decrease. Some parts of heart have a very high speed motion, and echocardiography videos, obtained by available systems can’t take enough number of frames to show them well. So, to achieve an echocardiography image set with a better visual quality, more frames are necessary between two frames at a great distance. Here, we use dimension reduction techniques to find out the number of suitable frames between two consecutive frames to show the fast motions better, but don’t take much time. We project images to a 3-dimentional space by this way. Greater difference between the frames, results greater distance between corresponding embedded points. Thus, the distance between the embedded points is a scale for the suitable number of frames, needed between two successive frames. These frames are produced with the registration techniques. On the other hand, heart doesn’t have a constant speed during a cycle, but echocardiography images are recorded with constant speed. So, frames at a greater distance show fast motions of the heart, and frames at a lower distance show slow motions of the heart. While, we put unequal number of frames between successive frames, and in this way remove temporal coordination of the image set. To solve this problem, we put efficient number of linear average of available frames, in places that the number of inserted frames in between available frames is less than maximum to obtain an equal number of frames between all successive frames.