Multivariate texture-based segmentation of remotely sensed imagery for
extraction of objects and their uncertainty
ARKO LUCIEER*{, ALFRED STEIN{ and PETER FISHER{
{International Institute for Geo-Information Science and Earth Observation (ITC),
Department of Earth Observation Science, PO Box 6, 7500 AA Enschede,
The Netherlands
{University of Leicester, Department of Geography, Leicester LE1 7RH, UK
In this study, a segmentation procedure is proposed, based on grey-level and
multivariate texture to extract spatial objects from an image scene. Object
uncertainty was quantified to identify transitions zones of objects with
indeterminate boundaries. The Local Binary Pattern (LBP) operator, modelling
texture, was integrated into a hierarchical splitting segmentation to identify
homogeneous texture regions in an image. We proposed a multivariate extension
of the standard univariate LBP operator to describe colour texture. The paper is
illustrated with two case studies. The first considers an image with a composite of
texture regions. The two LBP operators provided good segmentation results on
both grey-scale and colour textures, depicted by accuracy values of 96% and 98%,
respectively. The second case study involved segmentation of coastal land cover
objects from a multi-spectral Compact Airborne Spectral Imager (CASI) image,
of a coastal area in the UK. Segmentation based on the univariate LBP measure
provided unsatisfactory segmentation results from a single CASI band (70%
accuracy). A multivariate LBP-based segmentation of three CASI bands
improved segmentation results considerably (77% accuracy). Uncertainty values
for object building blocks provided valuable information for identification of
object transition zones. We conclude that the (multivariate) LBP texture model in
combination with a hierarchical splitting segmentation framework is suitable for
identifying objects and for quantifying their uncertainty.
1. Introduction
Geospatial data quality is a topic that has been covered frequently in recent
scientific literature on GIS and remote sensing (Foody and Atkinson 2002). An
important component of data quality is data uncertainty. Poor class definition,
gradual transition zones or fuzzy boundaries, mixed pixels, and incomplete or
imperfect data give rise to uncertainty in remotely sensed image-classification
results. Both fuzzy and probabilistic classification techniques can help to model and
quantify uncertainty. In recent years, much research has focused on modelling
uncertainty in remotely sensed image classification (Foody 1996, Hootsmans 1996,
Canters 1997, Fisher 1999, van der Wel 2000, Zhang and Foody 2001, Foody and
Atkinson 2002). It mainly focused on uncertainty of spectral classification on a
pixel-by-pixel basis. As such, it partially ignored potentially useful spatial relations
between pixels.
*Corresponding author. Email: arko@lucieer.net
International Journal of Remote Sensing
Vol. 26, No. 14, 20 July 2005, 2917–2936
International Journal of Remote Sensing
ISSN 0143-1161 print/ISSN 1366-5901 online # 2005 Taylor & Francis Group Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/01431160500057723

Object-oriented approaches to remotely sensed image processing have become
popular with the growing amount of high-resolution satellite and airborne imagery.
Segmentation techniques extract spatial objects from an image (Gorte and Stein
1998, Lucieer and Stein 2002). They extend classification, as spatial contiguity is
an explicit goal of segmentation, whereas it is only implicit in classification.
Uncertainty in a segmented or classified image can affect further image processing.
In particular, in areas where fuzzy objects or objects with indeterminate boundaries
dominate, an indication of segmentation uncertainty is important.
A straightforward approach to identify fuzzy objects is to apply a fuzzy c-means
(FCM) classification. This classifier gives membership values of belonging to each
class for each pixel. A thematic map can be obtained from this result by labelling the
pixels according to the class with the maximum membership value. However, pixel-
based classifiers, like the FCM, do not take spatial relations, also known as pattern
or texture, between pixels into account. We argue that a texture-based segmentation
approach (i.e. including the spatial component) can help to identify fuzzy objects.
Texture reflects the spatial structure of pixel values, and it is therefore indispensable
in segmenting an area into sensible geographical units.
Texture analysis has been addressed and successfully applied in remote sensing
studies in the past. An interesting overview paper concerning texture measures is
that of Randen and Husøy (1999). Recently, Ojala and co-workers have further
pursued an efficient implementation and application towards texture-based
segmentation (Ojala et al. 1996, 2002b; Ojala and Pietika¨inen 1999; Pietika¨inen
et al. 2000). Their Local Binary Pattern (LBP) measure is superior to most of the
traditional texture measures in segmentation of texture images (Ojala et al. 1996).
LBP is a rotation-invariant grey-scale texture measure.
The aim of this study is to develop and apply a supervised multivariate texture
segmentation technique to identify objects from remotely sensed imagery. It is
applied to an image with a texture composition and to an airborne multispectral
image of a coastal area in the Ainsdale Sands, northwest coast of England. It builds
on the work of Lucieer and Stein (2002) and Lucieer et al. (2004), and further
explores the use of multivariate texture. In addition, we focus on quantification of
object uncertainty to identify transition zones.
2. Methods
2.1 Texture
Image texture can provide valuable information for identification of objects. Not
only can the human visual system distinguish objects based on colour, but texture
plays an important role as well. A major characteristic of texture is the repetition of
a pattern or patterns over a region. The pattern may be repeated exactly, or as a set
of small variations, possibly as a function of position. There is also a random aspect
to texture, because size, shape, colour and orientation of pattern elements
(sometimes called textons) can vary over a region.
A comparative study of texture measures is given in Randen and Husøy (1999).
They conclude that a direction for future research is the development of powerful
texture measures that can be extracted and classified with a low computa-
tional complexity. A relatively new and simple texture model is the LBP (Pietika¨inen
et al. 2000, Ojala et al. 2002b). It is a theoretically simple yet efficient approach to
grey scale and rotation-invariant texture segmentation based on local binary
2918 A. Lucieer et al.