You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Mapping without the sun
Zhang, Jixian

LU Shuqiang
Beijing University of Civil Engineering and Architecture, Beijing, 100044, China
Key Word: Texture analysis, High-resolution RS image, Segmentation, Gray level co-occurrence matrix
Higher spatial resolution remote sensing (RS) images bring us more information and higher accuracy for our applications. At the
same time, it also brings new challenges in high resolution image processing. In order to increase the segmentation accuracy, a
texture feature based algorithm is proposed in this paper. This algorithm extracts five texture features from the original image. Two
different original images are chosen to exam the algorithm. Finally, the segmentation result is given. And from the results it can been
seen that this algorithm is effective for high spatial resolution RS images.
As the rapid development of the remote sensing (RS)
technology, the spatial resolution of acquired images has
reached 0.6m or higher level. It brings us more information and
higher accuracy for our applications. In addition to rich spectral
information, the structural, figure and textural features appear
more obvious in high resolution image(LIU, 2003).
As the result of prevailing of high resolution RS image, it is
easier to extract useful information artificially not only for
expert, but also for other person. However, it become more
difficult to realize automatic image processing for computer.
The reason is that most traditional methods focused only on
spectral information in RS images. The structural, figure and
textural information are hardly used in image processing. In
order to increase the accuracy and speed of image processing, it
is necessary to integrate other useful information in image
processing. In the field of high spatial resolution RS image
segmentation and classification, the structural and textural
features are commonly used.
Many texture analysis methods have been proposed in various
filed. Texture analysis methods could be classified as statistic
method, geometry method, model-based method and signal
process method. In the last decades, different texture analysis
methods have been applied to high spatial resolution RS image
processing. A RS image segmentation algorithm based on
simplified random field model was proposed in (Ming, 2004).
Texture analysis method based on wavelet transform has been
applied to high spatial resolution RS image
Classification(HUANG, 2006). A new texture method has been
proposed and applied to residential area half-automatic
extraction from high spatial resolution RS image(Su, 2004).
The key problem focused on how to extract useful information
efficiently from high resolution images.
displacement d, direction 0, and original gray level i,
P(i,j I d,9) re p resen t s the probability of gray level j
Many texture features can be defined according to gray level
co-occurrence matrix. Fourteen texture features have been
defined by Haralick. In this paper, five of them are chosen to
use in image segmentation. They are listed as the follows.
(1) Energy
(2) Entropy
A=~ZI 1 p Ai,№gPAU)
(3) Contrast
f,=YL^-n lp A‘ ! j
(4) Homogeneity
f = W_^M_
4 4-ri Hi-Jf
(5) Covariance
fs=Yull P d ( f . Jf 0' ~ /A )U ~ M y )
‘ j
p(iJ) = P(Uj I d,V) + P(iJ | ¿/,45°) + PjiJ | d,9(f) + P(iJ @135)
Px and are the means of P d( x ) and Pd 0 ; ) respectively.
P d {x)^P{x,j)
Gray level co-occurrence matrices suggested by Haralick have Before the segmentation, it should be supposed that the size of
become one of the most well-known and widely used texture any connected textur e area in RS images must be large than
features. Texture could be seen as the spatial distribution and
correlation between gray levels of image in this method. Given