Teaser: Ilustration of foreground/background sketch-based interactive mesh segmentation. Given a bunny model (a), the user marks a blue stroke on the area of the bunny¨s head (foreground) and a red stroke on the area of the bunny¨s body (background) and obtains a segmentation result in (b). The user then specifies an additional red stroke on the area of the bunny¨s body and the algorithm updates the segmentation using the new information, as shown in (c).
Abstract
This paper presents an
extensive comparative evaluation of five popular foreground/background
sketch-based interactive mesh segmentation algorithms, addressing the
quantitative assessment of the accuracy, efficiency, and stability of
each algorithm. To facilitate the comparison, we have developed a
complete framework with an intuitive and simple sketch-based interface
to enable interactive mesh segmentation by marking strokes to specify
the foreground and background with the mouse buttons, allowing us to
quantify the algorithms in a unified manner. The evaluation has been
performed via extensive user experiments in which each participant was
assigned to segment models with the evaluated algorithms and the
corresponding update of each segmentation was recorded as a new
refinement when additional interactions were added. We then collected
the segmentations from participants and evaluated them against the
ground truth corpus constructed from the Princeton segmentation
database. To investigate how well the interactive segmentations match
the ground-truth, five metrics were used to measure the boundary and
region accuracy of segmentations. By studying the experimental results,
we have analyzed the performance of the evaluated algorithms and
provided valuable insights into their characteristics.
Supplementary file (detail of user study) (2.0M)
Data download:
See project page of
Easy Mesh Cutting for all
the works of foreground/background sketch-based user interfaces for mesh
cutting.
Other sketch-based user interfaces for mesh cutting, see [Project page].
This work is supported by
the National Natural Science Foundation of China (61070071), the 973
National Key Basic Research Foundation of China (2009CB320801), and the
Fundamental Research Funds for the Central Universities (2010QNA3039).
title = {A Comparative Evaluation of Foreground/Background Sketch-based Mesh Segmentation Algorithms},
author = {Min Meng and Lubin Fan and Ligang Liu}
journal = {Computers & Graphics (Proceedings of Shape Modeling International)},
volume = {35},
number = {3},
pages = {650-660},
year = {2011}
}