Approaches to classification of multichannel images
CIARP'06 Proceedings of the 11th Iberoamerican conference on Progress in Pattern Recognition, Image Analysis and Applications
| Hi-index | 0.00 |
We present results on quantifying the exploitability of compressed hyperspectral remote sensing imagery. We investigate the performance of various feature extraction tasks on hyperspectral images coded using the JPEG-2000 Standard. Spectral decorrelation is applied prior to compression using the Karhunen-Loeve Transform (KLT) and 9-7 wavelet transform. We focus on the performance of several classification tasks: supervised, unsupervised and hybrid, and report quantitative results as functions of the compressed bit rates. We show that the feature extraction tasks examined in this paper can be performed with 99% accuracy at bit rates as low as 0.125 bit/pixel/band(bpppb). This strongly suggests that we need not limit remote sensing applications to lossless compression only, and we show that many common classification tools perform very reliably on images compressed/reconstructed at low bit rates.