Elsevier

Graphical Models

Volume 102, March 2019, Pages 10-18
Graphical Models

Portrait relief generation from 3D Object

https://doi.org/10.1016/j.gmod.2019.01.002Get rights and content

Abstract

Automatic generation of reliefs from 3D objects has received great attention in the past few years. The existing methods, which have been developed for general relief generation, are not suitable to handle portrait models that have distinct geometrical features. In this paper, we present a novel method for portrait relief modeling which takes mesh deformation as the theoretical basis. In a pre-processing stage, we segment the input model into several meaningful regions so that different portrait regions could be separately processed. To emphasize visually important features, we further compute object saliency and apply normal enhancement. Once the projective direction is selected by users, the input object is transformed into a compressed height field, followed by a Laplacian-based mesh deformation to optimize the height field. The fine details, salient features and depth structures could be well constructed in the final relief. Experiments and comparisons with state-of-the-art methods demonstrate the effectiveness of our method.

Introduction

Portrait relief has been widely used in a variety of media such as coins, medals, housewares and architecture. It is a sculpture form where the portrait elements remain attached to a solid background of the same material. Compared to the in-the-round sculpture, a portrait relief saves much more depth, but still gives the impression of 3D shape under illumination.

In real-world scenarios, there are three main types of portrait reliefs: high relief, middle relief and bas-relief. High relief is a sculpture that projects more than half of the figure, which may be in parts detached from the background. Mid-relief and bas-relief are sculptures that project only up to half of the portraits, and no elements are undercut or fully disengaged from the background, as shown in Fig. 1. This paper focuses on the generation of portrait mid-relief and bas-relief, where we represent the relief geometry as a 2.5D height field, a surface giving every point above the background a single height value.

Traditional portrait relief modeling relies on the skills of sculptors, which is time-consuming. Recently, automatic generation of reliefs from 3D objects has received great attention. As the input object is fully round in 3D space, users can freely change the projective direction, and generate a desired relief with the same pose. Before this work, a number of object-based methods have been proposed to generate 2.5D reliefs. These approaches are developed for general relief generation, but maybe not capable enough to handle portrait models that have varying geometrical features such as hairs, facial features and clothes.

In this paper, we present a novel method to generate reliefs from portrait models. Our goal is to construct a 2.5D height field that preserves or enhances the appearance of the portrait. Considering people are sensitive to the depth structure of human figure, we also expect that the depth structure of the portrait could well be constructed. To achieve this, we take advantage of Laplacian-based mesh deformation, which has proved an efficient way to generate bas-reliefs [1], [2]. Initially, we segment the input model into several meaningful components so that different portrait regions could be separately processed. Once the projective direction is selected by users, we transform the input portrait into a compressed height field, followed by an optimization procedure to enhance the portrait appearance. Experiments and comparisons with previous methods show that our method is effective in generating portrait reliefs with appealing appearance.

This paper is organized as follows. Related works are described in Section 2. Section 3 introduces technical details of the modeling pipeline. In Section 4, we present the experimental results and comparisons. Conclusions are given in Section 5.

Section snippets

Related works

Starting from the pioneer work of Cignoni et al. [3], automatic or semi-automatic relief modeling from a 3D object has been a subject of interest in computer graphics, which aims to compress the input object with no interactions or minor human efforts, while preserving the appearance of the input. The existing methods focus on two fundamental problems. One is to recover the geometrical details due to large compression of the height field. The other is to remove large height gaps (height

Overview of the pipeline

Our method for generating a portrait relief from 3D object consists of three main steps, shown in Fig. 2.

  • 1)

    Pre-processing. We segment the input model into several meaningful regions. We also enhance the surface normals, and define mesh saliency on the input model to capture visually important features.

  • 2)

    Height field construction and optimization. Once the projection direction is determined by users, we transform the input object into a compressed height field, together with transferred

Experiments and discussions

We implemented the proposed algorithm in C++ and tested it on a PC with an Intel(R) Core(TM) i5-3230M CPU @ 2.6 GHz and 4 GB of memory. In the pre-processing stage, it took about six seconds to segment the input model with 200k vertices. The saliency computation took about two minutes, and normal enhancement took about five seconds. Given a specified projective direction, the time for height field construction was about ten seconds, and it took another eight seconds to optimize the relief

Conclusions and future works

We present a novel method for portrait relief generation from 3D object in this paper. The main difference from previous works is that we handle portrait features in different manners. The relief is generated by a Laplacian-based mesh deformation. Fine details, visually salient features and depth structure of the input object could be well constructed in the final relief. Experiments and comparisons with state-of-the-art methods show the effectiveness of our method.

Currently, the evaluations of

Acknowledgement

The authors would like to thank the anonymous reviewers for their careful reviews and valuable suggestions. This work was supported in part by the National Natural Science Foundation of China (Grant No. 61772293, No. 61572161, No. 61602277), and the NSFC-Zhejiang Joint Fund of the Integration of Informatization and Industrialization (U1609218).

References (23)

  • Y. Zhang et al.

    Real-time bas-relief generation from a 3D Mesh

    Graphical Models

    (2013)
  • Y. Zhang et al.

    Line-based sunken relief generation from a 3D mesh

    Graph. Models

    (2013)
  • Z. Bian et al.

    Preserving detailed features in digital bas-relief making

    Comput. Aided Geometric Design

    (2011)
  • Y. Zhang et al.

    Bas-relief generation and shape editing through gradient-based mesh deformation

    IEEE Trans. Visualization Comput. Graphics

    (2015)
  • Y. Zhang et al.

    Adaptive Bas-relief generation from 3D object under Illumination

    Comput. Graphics Forum

    (2016)
  • P. Cignoni et al.

    Computer-assisted generation of bas- and high-reliefs

    J. Graphics Tools

    (1997)
  • T. Weyrich et al.

    Digital bas-relief from 3D scenes

    ACM Trans. Graphics

    (2007)
  • W. Song et al.

    Automatic generation of bas-reliefs from 3D shapes

  • J. Kerber et al.

    Feature sensitive bas-relief generation

  • X. Sun et al.

    Bas-relief generation using adaptive histogram equalization

    IEEE Trans. Visualization Comput. Graphics

    (2009)
  • Z. Ji et al.

    Bas-relief modeling from normal images with intuitive styles

    IEEE Trans. Visualization Comput. Graphics

    (2014)
  • Cited by (13)

    • ReliefNet: Fast Bas-relief Generation from 3D Scenes

      2021, CAD Computer Aided Design
      Citation Excerpt :

      Some of the proposed methods are designed for modeling generic types of bas-reliefs ([2; 3; 4; 5; 6; 7; 8; 9]). Other methods are designed specifically for modeling a certain type of bas-reliefs, for example, portrait bas-relief generation from a 3D object or a 2D image ([10; 11; 12]). Although compressed to a limited height interval, the relief models have rich geometric details.

    • Dynamic spline bas-relief modeling with isogeometric collocation method

      2020, Computer Aided Geometric Design
      Citation Excerpt :

      Digital bas-relief modeling has been studied a lot as digital models are convenient to design and modify. In the past years, researchers have proposed many methods to construct bas-reliefs from sketches, 2D images and 3D scenes (Kerber et al., 2012; Zhang et al., 2019a,b). But those methods only are appropriate for static bas-relief model construction, the dynamic bas-relief models which can be used in simulation, 3D animation and virtual reality are not considered yet.

    • Example-driven modeling of portrait bas-relief

      2020, Computer Aided Geometric Design
      Citation Excerpt :

      Therefore, we have a total of 84 feature points on the 3D mesh. Then, we detect the boundary vertices and use them as starting points to flood-fill each region following curvature-based flood-filling algorithm Zhang et al. (2019a). The segmentation results are shown in Fig. 3(b), in which each vertex on mesh belongs to a unique region with a few user interactions.

    • Neural Modeling of Portrait Bas-Relief from a Single Photograph

      2023, IEEE Transactions on Visualization and Computer Graphics
    View all citing articles on Scopus
    View full text