104
Figure 16. The result of classification using contrast
texture feature image assistant spectrum feature
Figure 17. The result of classification using second
moment texture feature image assistant spectrum feature
Figure 18. The result of classification using entropy
texture feature image assistant spectrum feature
Precision
(Percent)
Kappa
The maximum likelihood
classification based on spectrum
feature
84.0789
0.782
8
The maximum likelihood
classification of entropy image
assistant
92.7664
0.906
4
The maximum likelihood
classification of second moment
image assistant
92.5438
0.893
7
The maximum likelihood
classification of contrast image
assistant
93.2179
0.919
5
The maximum likelihood
classification of correlation
image assistant
91.9348
0.893
5
Table 2. The statistic data of classification precision
The classification result appraisal and compares:
From above-mentioned experiment result, we can draw the
following conclusions:
1) when the texture features were employed in the process of
image classification , the general classification accuracy
reached 92.6%, without the texture feature auxiliary, the
general classification accuracy was only 84.07%.
2) Using above four kind of textures feature to participate in the
classification, we found the inhabited area, the forest, the
badlands and the paddies had improvements in classified effect.
The contrast image assistance classification was used to
improve on classification effects of the water system. Flowever,
compared with original image, it still had the phenomenon of
divides by mistake. Classification effects of other several kinds
of texture feature images for water system were not good,
which caused many divides by mistake.
3) Looked through the effect of above several features images
assistance classification, the clarity and the dividing degree had
certain enhancement to some extent, we was allowed to see
from the statistical result of classification that the precision and
the kappa coefficient all has the certain improvement; But
phenomenon of wrong dividing and leaks dividing still exist.
5. CONCLUSIONS
This article mainly studies the algorithm of texture feature
image extraction based on the gray level co-occurrence matrix
and the precision appraisal. The program of extraction texture
feature image was realized by VC++. We extract four texture
images by this program, that is, second moment image, contrast
image, correlation image and entropy image. Next, we take
these four texture feature as a band respectively to combine
with three bands of spectrum feature, which will form a image
with 4 bands and was finally classified by means of the
algorithm of the maximum likelihood classification. Indicated
through the experiment, supervisor classification method of
texture feature assistance spectrum feature classification
improves the precision classification to some extent, but
phenomenon of wrong dividing and leaks dividing still exist.
REFERENCES
[1] Jia Yong-hong, K, 2003. Processing of Remote Sensing
Imag, Wuhan University Press.
[2] C. S. Lu, P. C. Chung and C. F, 1997. Chen. Unsupervised
texture segmentation via wavelet transform. Pattern
Recognition, 30 (5).
[3] Guo De-jun. Song Zhe-cun, 2005. A Study on Texture Im
age Classifying Based on Gray-level Co-occurrence Matrix,
FORESTRY MACHINERY & WOODWORKING
EQUIPMENT(In Chinese;, 33(7), pp. 21-23.
[4] Zhao Hong-rui, YAN Guang-jian, Deng Xiao-lian, Wang
Jin-di, YANG Hua, LI Xiao-wen, 2003. A Classification
Method Based on Spatial Information, JOURNAL OF
REMOTE SENSING, 17(5).
[5] Di K. C., Li D. R., Li D. Y„ 2000. Study Of Remote
Sensing Image Classification Based On Spatial Data Mining
Techniques, Journal of Wuhan Technical University of survey
ing and Mapping, 25, (1), pp. 42-48.
