Feature Detection
Contributors
- Vasyl Mykhalchuk, Frederic Cordier, Hyewon Seo (University of Strasbourg)
- Guillaume Lavoué (LIRIS, Lyon), Frederic Cordier (LMIA, Mulhouse), Chaker Larabi (XLIM, Poitier)
Abstract
We have developed a method for the extraction of spatio-temporal (dynamic) feature points from animated meshes. To the best of our knowledge, we are the first who propose the spatio-temporal feature detector in animated meshes. Apart from guiding the registration of deforming meshes in our later work, the results of our dynamic feature detection algorithm serve as a starting point in a number of promising applications, such as animation temporal alignment, animated mesh simplification, viewpoint selection etc. Our two main contributions are: a multi-scale representation (Gaussian pyramid) of the surface deformation characteristics and a novel spatio-temporal Difference of Gaussians (DoG) filter for deforming meshes. Our algorithm demonstrates a highly desirable property of consistent feature point detection across semantically similar deforming shapes, as illustrated in the image above.
Another work related to these is the validation of computational mesh saliency by using an eye-tracker, which has been the result of collaboration with Guillaume Lavoué (LIRIS), Frederic Cordier (LMIA), and Chaker Larabi (XLIM). In search of validating our dynamic feature extraction work, we have set our goal to build the ground truth by analyzing the human visual saliency using an eye-tracker. As a start, we have decided to focus on the human visual saliency analysis on the 3D shapes, which we plan to extend to the shapes with appearance and with deformation.
Publications
Lavoué G., Cordier F., Seo H., Larabi M., "Visual Attention for Rendered 3D Shapes". Comput. Graph. Forum 37(2): 191-203 (2018).
[Website]
