145.

[Cover] [Contents] [Index]

Page 228

putational cost is to reduce image grey levels. A simple method known as equal-probability quantising algorithm (Haralick, 1973) can be used to achieve this end. Another method called the linked-list algorithm (Clausi and Jernigan, 1998), in which GLCM texture features are computed by the use of linked lists, is also found to be successful in increasing the computational efficiency of the GLCM texture feature process.

Finally, with respect to frequency domain filtering, the most successful texture segmentation result is derived from the combination of ring range [5, 7] and both horizontal and vertical wedges. One drawback, mentioned previously, is that the optimal ring range is hard to define, thus making segmentation results more unstable.

5.6.4 Texture measure of remote sensing patterns

The texture quantisation approaches described above are next applied to real remotely sensed patterns. Figure 5.24 shows a test image of size 256 ×256, which is made up be four 128×128 subimages extracted from different areas of the SIR-C SAR image shown in Figure 5.10b. The corresponding legends of these subimages (from upper left to lower right) are granitic, intrusives, volcano-sedimentary and alluvium, respectively. The subimages reveal different land surface structures. The MAR model, GLCM, frequency domain filtering, and one of multifractal method (only Equation (5.32) is tested) with the same model parameters are again used to quantify these four texture patterns. Each subimage is further divided into sixteen blocks, each 32×32 pixels in size. Classification results are shown in Figure 5.25. Both MAR and GLCM show relatively good texture discrimination ability compared with the other methods.

Figure 5.24 Test image made up by four different lithological types extracted from a SIRC SAR image. From upper left to lower right, clockwise: granitic, intrusives, volcano sediment and alluvium. SIR-C data provided by NASA Jet Propulsion Laboratory, Pasadena, CA, USA.

[Cover] [Contents] [Index]


Classification Methods for Remotely Sensed Data
Classification Methods for Remotely Sensed Data, Second Edition
ISBN: 1420090720
EAN: 2147483647
Year: 2001
Pages: 354

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net