Information Technology Reference
In-Depth Information
Chapter 12
Synthesizing Natural Images Using Spatial
Layout Information
Ling Shao and Ruoyun Gao
*
Abstract.
We propose an algorithm for synthesizing natural images consisting of
composite textures in this chapter. Existing texture synthesis techniques usually fail
to preserve the structural layout of the input image. To overcome this drawback, a
target image which contains the layout information of the input is used to control
the synthesis process. With the guidance of this target image, the synthesized output
resembles the input globally and is composed of the original textures. For images
composed of textured background and a non-textured foreground object,
segmentation is first applied on the main object and texture synthesis is used on the
background texture. In comparison to other synthesis methods, the proposed
solution yields significantly better results for natural images.
1 Introduction
Texture synthesis has experienced great development in the past two decades, and
manifested its significance in rendering synthetic images. By reproducing visual
realism of the physical world, it is used to create texture of any size. The goal of the
texture synthesis can be stated as follows: given a texture sample, synthesize a new
texture that, when perceived by a human observer, appears to be generated by the
same underlying stochastic process [4].
Texture synthesis techniques are mainly divided into two groups: parametric and
non-parametric. Parametric texture synthesis usually only works for simple and
regular textures. Non-parametric or example-based texture synthesis can deal with a
wide variety of texture types, from regular to stochastic. Among example-based
