17-03-2014, 10:15 PM
Abstract—this paper presents a simple approach for adding
more details to non-photorealistic image without increasing the
complexity of the model. Bump mapping technique is
introduced to create the details that can vary automatically
with the change of light and view direction, but this technique
originally aims to make a rendered surface look more realistic,
so we present an algorithm to stylize it the same to achieve
overall stylistic effect. We also present a novel approach for
drawing multiple lines in GPU, and all the algorithms in this
paper are based on GPU.
Keywords- Bump mapping, GPU, non-photorealistic, NPR
I. INTRODUCTION
Non-photorealistic rendering (NPR) is an important
domain of computer graphics and develops rapidly in the
past decade. Compared with photorealistic rendering
techniques in which researchers have focused on expressing
objects as it is, NPR techniques emphasize artistic, sensual
side [7]. It breaks free from the constraint of photorealism
that many other rendering techniques strive for and generates
images and animations that at least in some parts appear to
be made by hand [8]. NPR has been studied variously on oil
painting, water color, pen and cartoon, and much research
focuses on producing stylistic image in level of geometric
model, stylized textures are used to reflect the details on the
model, but they all work just like static pictures sticking to
flat surfaces, there is no interaction with the lights in the
environment. In photorealistic rendering, bump mapping
technique [9] is a good candidate to generate more details on
the surface of model without using more polygons, which is
advantageous especially for a real-time system. Since NPR
aims to create image that appears to be unrealistic, in this
paper, we present approach to introduce bump mapping into
NPR by stylizing the bump mapping at run time. In this
paper, we implement our algorithm to extract the outline in
texture level, simulate the pencil sketch as well as cartoon
style, all of the implementation are based on GPU.
II. RELATED WORKS
We first survey some of related works here.
Bump mapping was originally introduced by Blinn [12]
in 1978. The implementations using graphics hardware have
been proposed more recently, and dot-product bump
mapping is preferred nowadays for its better result and
flexibility. HEIDRICH. W. and SEIDEL H. [13] directly
stores the normals of the surface in texture maps, Olano and
Lastra [14] used a more general approach to built graphics
hardware that can run small shading programs and hence is
capable of doing bump mapping.
Georges Winkenbach and David H.Salesin[4] create a
system that takes polygonal models as input and renders
them in pen-and-ink style. Jodoin et al. focus on hatching
and approach the problem as a kind of texture synthesis
[10]. The interactive drawing system of Salisbury et al.
allows users to select from stored stroke textures or to create
them[11].
For a survey about silhouette detection, see Wang Ao-yu
et al[15], much of the work addresses the outline detection
in level of 3D Mesh, in our algorithm, we extent it to level
of texture, the inspiration is still from [17] edge detection in
image space.
more details to non-photorealistic image without increasing the
complexity of the model. Bump mapping technique is
introduced to create the details that can vary automatically
with the change of light and view direction, but this technique
originally aims to make a rendered surface look more realistic,
so we present an algorithm to stylize it the same to achieve
overall stylistic effect. We also present a novel approach for
drawing multiple lines in GPU, and all the algorithms in this
paper are based on GPU.
Keywords- Bump mapping, GPU, non-photorealistic, NPR
I. INTRODUCTION
Non-photorealistic rendering (NPR) is an important
domain of computer graphics and develops rapidly in the
past decade. Compared with photorealistic rendering
techniques in which researchers have focused on expressing
objects as it is, NPR techniques emphasize artistic, sensual
side [7]. It breaks free from the constraint of photorealism
that many other rendering techniques strive for and generates
images and animations that at least in some parts appear to
be made by hand [8]. NPR has been studied variously on oil
painting, water color, pen and cartoon, and much research
focuses on producing stylistic image in level of geometric
model, stylized textures are used to reflect the details on the
model, but they all work just like static pictures sticking to
flat surfaces, there is no interaction with the lights in the
environment. In photorealistic rendering, bump mapping
technique [9] is a good candidate to generate more details on
the surface of model without using more polygons, which is
advantageous especially for a real-time system. Since NPR
aims to create image that appears to be unrealistic, in this
paper, we present approach to introduce bump mapping into
NPR by stylizing the bump mapping at run time. In this
paper, we implement our algorithm to extract the outline in
texture level, simulate the pencil sketch as well as cartoon
style, all of the implementation are based on GPU.
II. RELATED WORKS
We first survey some of related works here.
Bump mapping was originally introduced by Blinn [12]
in 1978. The implementations using graphics hardware have
been proposed more recently, and dot-product bump
mapping is preferred nowadays for its better result and
flexibility. HEIDRICH. W. and SEIDEL H. [13] directly
stores the normals of the surface in texture maps, Olano and
Lastra [14] used a more general approach to built graphics
hardware that can run small shading programs and hence is
capable of doing bump mapping.
Georges Winkenbach and David H.Salesin[4] create a
system that takes polygonal models as input and renders
them in pen-and-ink style. Jodoin et al. focus on hatching
and approach the problem as a kind of texture synthesis
[10]. The interactive drawing system of Salisbury et al.
allows users to select from stored stroke textures or to create
them[11].
For a survey about silhouette detection, see Wang Ao-yu
et al[15], much of the work addresses the outline detection
in level of 3D Mesh, in our algorithm, we extent it to level
of texture, the inspiration is still from [17] edge detection in
image space.