An Efficient Warp-Based Motion Magnification Method to Reveal Subtle Changes in Video

Abstract

Video motion magnification can amplify subtle motions and even reveal small color changes in video. Ordinary methods analyze the signal change at each pixel over time at different spatial scales and orientations. These methods inevitably amplify the noise and cause ringing artifacts in video. State-of-the-art methods relying on filters via learning also produce excessive blurring in images. In this paper, we present a warp-based video motion magnification method in which only one-frame latency is maintained. We propose a Lagrangian motion magnification method, which involves image deformation and optical flow techniques. Motion magnification is achieved by warping the video frame. The method is guided by feature points and only uses previous motion, while simultaneously maintaining the original video details without noise amplification. With this approach, the proposed method can work online in real time. Experimental results show that our method can achieve high-quality results and significantly reduce artifacts, compared with state-of-the-art techniques