rer 
Id height 
ause the 
pervised 
ification 
d for an 
usly, the 
ly dete 
in poor 
> of the 
^, On the 
. used in 
AP laser 
.. 339. 
ta. 
'osium 
ISPRS 
). Final 
of 
n 
xture 
Stephane Paquis 
ROAD SURFACE TEXTURES CLASSIFICATION USING OPENING-BASED IMAGE 
PROCESSING 
Stéphane PAQUIS!, Vincent LEGEAY! and Hubert KONIK? 
!Lab. Central des Ponts et Chaussées - Nantes 
BP 4129 
44341 Bouguenais cedex, France 
stephane.paquisQ lcpc.fr 
?Lab. Ingénierie de la Vision 
Université Jean Monnet 
BP 505 
42007 Saint-Étienne cedex 01, France 
KEY WORDS: texture analysis, road surface inspection, hierarchical process, multiresolution cooccurrence matrix, mor- 
phological pyramid 
ABSTRACT 
This paper deals with an innovative approach for achieving an automatic vision system for road surface texture classifi- 
cation. A road surface is composed by aggregates with a particular grain size distribution and the mortar matrix. From a 
vision point of view, road surface images can be described as a set of objects of high intensities puts on a low intensity 
uniform background. In image processing, mathematical morphology provides a set of tools used to compare parts of 
an image with structuring elements of various sizes and shapes. Information about objects can be obtained by applying 
successives openings with structuring elements of increasing sizes. Our aim is to characterize 4 road surface textures with 
different size and shape distributions. In order to avoid the choice of a set of structuring elements, we define an opening 
transformation based on quadtree approach. At each step, each non overlapping 2 x 2 blocks take one new grey level, 
corresponding to its lowest one. Variations of texture observed at each step are studied through an original cooccurrence 
matrix. Such a matrix is computed between two consecutive pyramid levels. Classification is carried out by extracting 
textural features from the set of matrices. 
1 INTRODUCTION 
Our work aim is to develop an image analysis method for road surface classification. Future tasks will be to locate texture 
inhomogeneities relative to pavement surface defects such as cracks. Road surface texture images can be described as 
objects of different size and shape put on an uniform background. In France, 98% of pavements are hydrocarbon concrete 
which can be divided into two categories : surface dressing (chipping on binder layer) and bituminous asphalt (mixtures of 
coarse aggregate, fine aggregate with or without filler, and hydrocarbon binder). There are several surface dressing (SD) 
conceptions but from a surface aspect point of view surface texture has a strong roughness level because of aggregates 
angularity ; however 3 classes of bituminous asphalt can be made considering their percentage of voids : dense, semi- 
dense and open, whose principal members are respectively classical bituminous asphalt (CBA), ultra-thin bituminous 
asphalt (UTBA) and porous asphalt (PA). Of course in a same pavement classe several grain size distributions can occur. 
An example of studied pavements is given in figure 1. 
Multi-resolution pyramids are frequently used for detection and identification of objects or features of different size 
(Rosenfeld and Sher, 1988, Konik et al., 1993, Segall et al., 1999). In a general way, an images pyramid represents 
a collection of scene representation corresponding to an original image taken at low resolution. The data structure is 
initialised by placing the input image at the lowest level. The next levels are constructed by successively filtering and 
subsampling. The process finishes at the apex of the pyramid structure, when the last level is nonempty. Construction of 
images pyramid typically apply linear low- or pass-band filters. This technique removes all high frequency signal content, 
which alters the objects intensities and therefore induces region merging between successive levels. It is not suitable 
for tasks involving precise measurement of object size and shape. Morphological filters remove small features without 
altering high frequency information. These filters preserve essentiel shape characteristics while eliminating irrevelancies. 
By considering the nature of the road surface images, it is natural to use mathematical morphology, which deals directly 
with shape, as a basis for the construction of pyramid levels. 
This paper proposes to apply opening-based processing to road surface classification problem. The outline of this paper 
Is organized as follows : Section 2 reviews briefly mathematical morphological tools used to constructed morphological 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 685