Ammatzia Peled
30
25
20 +
15
10
© e CN co st © ©
— e s © co o
112 -
‘ Figure 3: Histograms of gray level values for the pixels within “Stone-Paved” roads.
2.2.3 Buildings & Roads Combined Together
According to the classification processing, of buildings and roads, separately, the results show a mixed detection of both
buildings and roads. Therefore, the decision was to implement an integrated classification. Also here, the results show
mixed detection, of buildings and roads, with a large percentage of "noise".
2.3 Conclusions
According to the above mentioned experiments, the conclusion was that using, only, radiometric data (from black and
white aerial photographs), is not enough to enable change identification, obviously. Therefore, the decision was to
continue the research by using color photographs and multi-spectral remotely sensed images. Nevertheless, the research
was continued to test “texture classification approach".
3 TEXTURE CLASSIFICATION
The objective of texture classification, is to determine specific pattern templates, which define, uniquely, a required
feature class. In this research, advanced algorithms for texture identifications in a statistical approach, were applied
Three types of statistical textures were tested:
(a) The differences between the gray level value of the tested pixel and the average value for the neighboring pixels.
(b) The maximum difference between the gray level value of the tested pixel and each of neighboring pixel.
(c) The difference between the extremum gray level values of the neighboring pixels.
3.1 Experiments
In the first step, a different "texture images" (according to the above three types) were produced according to different
filter sizes. The results of these “texture images” were detection of edges. The texture classification was developed ina
similar way to the GIS-Driven classification. For each feature (building or road), three histograms, which describe the
distribution of the above three types of statistical texture parameters, were generated. According to these histograms, the
groups of gray level values were determined, depending on the peaks and spikes depicted in the histogram functions.
For each group, an “estimated” frequency percentage of the texture value ranges was determined. This, according to the
mean and standard deviation of all histograms related to same feature class type. The statistical matching algorithms
were implemented for each pixel to one of the feature classes. Several texture classifications processes Wer
implemented, with a different filter size.
3.2 Conclusions
The results of the texture classification experiments, show a distinct detection of edges (for both roads and buildings)
.e., instead of detecting the feature itself, the operator detected it's edge. In spite of this successful edge-detection. I
was found that it is not possible to define the feature types by unique values set of the texture parameters.
716 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
Acco
white
concl
a geo
REF!
a
2-8 ee ME a N a eed
— RR.
ad en