A clustering-based method to estimate saliency in 3D animated meshes

• We present a new saliency estimation method for 3D animated mesh models.
• We present a motion based mesh segmentation technique that partitions the vertices in an animated mesh to logical clusters.
• The proposed model regards each cluster as a separate item.
• We verified the model׳s performance with an eye tracking based formal user study.
• Decreasing the calculation scope from thousands of vertices to tens of clusters provides significantly faster saliency calculation.

We present a model to determine the perceptually significant elements in animated 3D scenes using a motion-saliency method. Our model clusters vertices with similar motion-related behaviors. To find these similarities, for each frame of an animated mesh sequence, vertices׳ motion properties are analyzed and clustered using a Gestalt approach. Each cluster is analyzed as a single unit and representative vertices of each cluster are used to extract the motion-saliency values of each group. We evaluate our method by performing an eye-tracker-based user study in which we analyze observers׳ reactions to vertices with high and low saliencies. The experiment results verify that our proposed model correctly detects the regions of interest in each frame of an animated mesh.

Figure optionsDownload high-quality image (110 K)Download as PowerPoint slide