Joint entropy-based motion segmentation for 3D animations

Abstract

With the recent advancement of data acquisition techniques, 3D animation is becoming a new challenging subject for data processing. In this paper, we present a joint entropy-based key-frame extraction method, which further derives a motion segmentation method for 3D animations. We start by applying an existing deformation-based feature descriptor to measure the degree of deformation of each triangle within each frame, from which we compute the statistical joint probability distribution of triangles’ deformation between two consecutive subsequences of frames with a fixed length. Then, we further compute joint entropy between the two subsequences. This allows us to extract key-frames by taking the local maximal from the joint entropy curve (or energy curve) of a given 3D animation. Furthermore, we classify the extracted key-frames by grouping the key-frames with similar motions into the same cluster. Finally, we compute a boundary frame between each of the two neighboring frames with different motions, which is achieved by minimizing the variance of energy between the two motions. The experimental results show that our method successfully extracts representative key-frames of different motions, and the comparisons with existing methods show the effectiveness and the efficiency of our method.