Vis Comput (2016) 32:1537–1548
DOI 10.1007/s00371-015-1138-3
ORIGINAL ARTICLE
Two-level joint local laplacian texture filtering
Hui Du
1
· Xiaogang Jin
2
· Philip J. Willis
3
Published online: 29 May 2015
© Springer-Verlag Berlin Heidelberg 2015
Abstract Extracting the structure component from an
image with textures is a challenging problem. This paper
presents a novel structure-preserving texture-filtering appro-
ach based on the two-level local Laplacian filter. The new
texture-filtering method is developed by introducing local
Laplacian filters into the joint filtering. Our study shows that
local Laplacian filters can also be used for texture smoothing
by defining a special remapping function, which is closely
related to joint bilateral filtering. This finding leads to a
variant of the joint bilateral filter, which produces smooth
edges while preserving color variations. Our filter shares sim-
ilar advantages with the joint bilateral filter, such as being
simple to implement and easy to understand. Experiments
demonstrate that the new filter can produce satisfactory filter-
ing results with the properties of texture smoothing, smooth
edges, and edge shape preserving. We compare our method
with the state-of-the-art methods to demonstrate its improve-
Electronic supplementary material The online version of this
article (doi:10.1007/s00371-015-1138-3) contains supplementary
material, which is available to authorized users.
B
Hui Du
duhui@zjicm.edu.cn
B
Xiaogang Jin
jin@cad.zju.edu.cn
Philip J. Willis
P.J.Willis@bath.ac.uk
1
New Media College, Zhejiang University of Media and
Communications, Hangzhou 310018, China
2
State Key Lab of CAD&CG, Zhejiang University,
Hangzhou 310058, China
3
Department of Computer Science, University of Bath,
Bath BA2 7AY, UK
ments, and apply this filter to a variety of image-editing
applications.
Keywords Texture filtering · Structure extraction ·
Local Laplacian filters · Guidance image
1 Introduction
Edge-preserving image smoothing is very important and use-
ful in many image applications, e.g., detail manipulation,
abstraction, tone mapping, and image composition. It seeks
to decompose an image into structure and detail compo-
nents. Most existing work, such as the bilateral filter [25],
weighted least squares filtering [8], L
0
smoothing [28] and
local Laplacian filters [17], focused on separating structures
from details while being edge preserving. These algorithms
usually depend on gradient magnitude, pixel intensity and
Laplacian pyramid coefficients to obtain satisfactory edge-
preserving filtering results. While humans easily recognize
the structures of an image which contains rich textures, it is
difficult for a computer to automatically extract structures by
texture smoothing. Most edge-preserving filtering methods
do not explicitly address the structure-preserving texture-
smoothing problem. Textures may be considered as strong
edges if we apply these approaches straightforwardly to tex-
ture smoothing. As a result, textures are preserved instead of
being smoothed and unsatisfactory results may arise.
To solve this challenging problem, previous methods [4,6,
12,23,24,29,31] try to suppress textures to obtain structure-
preserving texture-smoothing results of images using differ-
ent strategies, such as the weighted least squares optimization
or the joint bilateral filtering schemes. These solutions can
achieve good separation results. However, rich color varia-
tions of the input may be flattened somewhat or structure
edges of the input may be damaged by these filters. For
123