89% correct
81% correct
Terrain categorization is an important example of machine exploitation using multispectral images. Based on pattern recognition techniques, the decompressed image pixels are classified into groups representing different terrain and land covers. This image shows the effects of compression artifacts on classification errors for a five-band subset of a Landsat scene. Clearly, the image compressed by modulated lapped transform yields a better categorization score than the JPEG.
|
|
|
| Original | MLT 8 to 1 89% correct | JPEG DCT 8 to 1 81% correct |
|
|
|