4.3 Image Enhancement & Vegetation Mapping
To acquire the maximum spectral variation of the IR image, Decorrelation Stretch was performed Stretch was performed. In the process of carrying out the Decorrelation Stretch, Principal Components (PC) 1,2 & 3 were created. The multi-variate statistical analysis of Principal Component Analysis map co-variances of the input three bands into new axes which are orthogonal to each other (Richards, 1995). Since the first Principal Component has the most variances of the three input bands, with reducing co-variances in lower components, the last component (PC3) has the most noise. The RGB composite of PC123 revealed the vignetting effect (Fig. 5) and the traverses of the PC1,2 & 3 (Fig.6) suggest that the noise caused by vignetting effect has been accumulated in PC3.


Figure 5. RGB composite of PC123 of original IR, VisGreen bands IR photo. (A1-B1 is the traverse line of Figure 6).

Figure 6. Digital Number (DN) values of PC1, 2 & 3 of traverse A1-B1 on Figure 5.
Using the anti-vignetting formula of ER Mapper, the vignetting effect accumulated in PC3 was corrected. The Decorrelation Stretch was performed on the PC1, 2 and vignetting effect corrected PC3. Then the Decorrelation Stretched PC1, 2& 3 were inverted back to the orginal space of the NIR, VisRed and VisBlue bands. The RGB composite of the Decorrelation Stretched 3 bands provide much more spectral variation (Fig.7) than the original RGB composite of the IR Airphoto (Fig. 2).


Figure 7. RGB composite of enhanced IR, VisRed, VisGreen bands IR photo.
Judging from the Eigenvector loading of Bands 1,2,3 of the IR Airphoto into PC1, 2, 3, the highest variation between the loading of the three bands is an PC number 2 (Table 1). Since the distinctive high NIR reflectance of vegetation is in band1 and that its loading is negative, PC2 is inverted to highlight vegetation as bright areas (Fig.8).


Figure8. Inverted PC2 of the IR photo, Highlighting vegetation.
Table 1. Statistics of the IR data

Covariance Eigenvectors	PC1	PC2	PC3
Band1 (NIR)	0.337	-0.689	-0.641
Band2 (Visible Red)	0.447	-0.483	0.753
Band3 (Visible Green)	0.829	0.540	-0.145

Healthy vegetation has distinctive high spectral reflectance in NIR and chlorophyll absorption feature in Visible Red. To discriminate vegetation from other surface cover types such as soil and water, a band ratio of band1/band2 (NIR/VisRed) was carried out. The band ratio highlighted the vegetation in bright areas (Fig.9).


Figure 9. Raito Band IR/VisRed of the IR photo, highlighting vegetation.