IMAGE SEGMENTATION FROM TEXTURE MEASUREMENT 
Dong-Cheon Lee 
Toni Schenk 
Department of Geodetic Science and Surveying 
The Ohio State University 
Columbus, OH 43210-1247 
U.S. A. 
Commission III 
ABSTRACT 
It is known that the human visual system, unsurpassed in its ability to reconstruct surfaces, employs different cues 
to solve this difficult task. The prevailing method in digital photogrammetry is stereopsis. However, texture may 
provide valuable information about the shape of surfaces. In this paper we employ Laws' method of texture 
energy transforms to extract texture information from digital aerial imagery. The images are convolved with 
micro-texture filters to obtain local texture properties. Each micro-texture feature plane is transformed into an 
texture energy image by moving-window to render macro-texture features. Finally, the macro-texture feature 
planes are combined and then clustered into regions of similar texture pattern. The method is implemented in a 
scale-space approach, and the boundaries obtained from texture are compared with physical boundaries of the 
image. 
KEY WORDS: Texture primitive, Micro-texture, Macro-texture, Texture energy, Image Analysis. 
1. INTRODUCTION 
The goal of digital photogrammetry is to reconstruct 
surfaces automatically. Surface reconstruction from 
raw imagery is known as an ill-posed problem. To 
solve this difficult task, different cues which 
contribute to object recognition and scene 
interpretation are employed. One of the important 
cues is texture. Texture may provide information to 
estimate shape, surface orientation, depth changes, 
material of objects. Texture information aids image 
analysis and interpretation. 
Many texture analysis methods have been developed 
during the last two decades. Among the great variety 
of available methods, Laws' approach of texture 
energy measures appears to be a suitable method 
(Ballard and Brown, 1982; Gool et af, 1985; Gong 
and Huang, 1988; Unser and Eden, 1990). 
Furthermore, this method resembles human visual 
processing of texture according to Laws' dissertation. 
One of the advantages of this method is to provide 
several texture feature planes from an original image. 
This is a great benefit especially if only monochrome 
imagery is available because to extract useful texture 
information from raw monochrome images is a 
difficult task even for the human vision system. More 
useful information and segmentation results could be 
obtained by integrating the additional texture feature 
planes. 
2. CHARACTERISTICS OF TEXTURE 
Texture is qualitatively described by its coarseness 
under the same viewing condition, and related to the 
repetition of the local spatial patterns. In addition to 
coarseness, other textural dimensions or parameters 
are commonly proposed, namely, contrast, density, 
roughness, directionality, frequency, regularity, 
uniformity, orientation, and so on (Tamura et af, 
1978). 
Texture is a sophisticate visual primitive since texture 
element (texel) is determined by contextual process 
and a different level of hierarchy. Texture primitives 
consists of micro-texture and macro-texture. Micro- 
texture is the smallest primitive while macro-texture is 
referred to larger primitive, i.e., macro-texture is 
homogeneous aggregation of micro-texture. These 
two primitives cannot be confused with fine texture 
and coarse texture. The coarseness of texture is 
related to the spatial repetition period of the local 
structure. Therefore, micro-texture and macro-texture 
are not related the coarseness. However, in fact there 
are not clear criteria to differentiate micro-texture 
from macro-texture primitives, rather it is related to 
somewhat psychological effect as well as image scale 
and resolution. Since texture is hierarchical, texture 
within texture primitives themselves is visible (Gool 
et al, 1985). It is important to understand how the 
human visual system works for texture discrimination 
195