Automatic land classification vs. data compression: a comparative evaluation

Presents an accurate comparison of three known lossy compression techniques. Already well established is the vector quantization and JPEG while the coding of transformed wavelet coefficients is a more recent technique. Multispectral remotely sensed images (Thematic Mapper) have been transformed by the Karhunen Loeve transform (KLT) before compression by the algorithms. As the goal of this accurate analysis is the compression for automatic classification of the images, the efficiency and quality of the techniques was evaluated by a supervised classification of the decoded images, the well known algorithm K-NN (k-nearest-neighbor) for remote sensing applications while the MSE (mean square error) was compared for visual aspects. The main goal of the compression of remotely sensed images is a reduction of the huge requirements for downlink and storage which are needed nowadays to allocate the very high and increasing quantity of data from recent sensors. The Karhunen Loeve transform first removes the interband correlation to produce the principal components of the images which are then compressed by the three principal algorithms. The obtained results of these particular and accurate analysis of the current compression techniques are quite surprisingly compared to other recent works where most compression techniques perform well for visual aspects (e.g. browsing)