18.3.1 Effect of Textural Enhancement on Image Classification
Figure 18.5 shows the original ortho-aerial photograph and enhanced images (the original photograph combined with the textural images). The MLC of supervised classification was adopted to discuss the benefit of this approach. Some environmental effects, such as shadow, aerosol, and terrain, create spectral variance and impact the accuracy of image classification. The spectral variance can not be corrected and reduced completely, even with the rapid development of the RS technique. Our results demonstrated that the enhanced image improved the classification results in shadowy regions; that is, the effect of image tone combined with texture is reduced shadow and terrain effect, although this approach did not work on nonreflective regions. The improvement in classification is shown in Fig. 18.6. Figure 18.7 illustrates the field survey data.
A total of 256 random points were evaluated, and the overall kappa value was 0.85 (Table 18.1). The kappa coefficients for forest, A. formosana, basin, and bare soil were 0.85, 0.81,0.92, and 0.82, respectively. According to the study by Janssen and Vanderwel in 1994, the kappa value should be higher than 0.7. Therefore, the classification results could be applied to the analysis of environmental factors, allowing us to gain niche information.