Non-negative bases in spectral image archiving

This thesis supposes an application of Principal Component Analysis (PCA), Non-negative Matrix Factorization (NMF) and Non-negative Tensor Factorization (NTF) for digital image archiving. It is aimed to develop new efficient methods for spectral image acquisition, compression and retrieval. It hypothesizes that the non-negative bases are more suitable for spectral archiving beside convenient orthogonal. The thesis introduces three fundamental components of the digital image archiving system. It gives an overview of the methods that were developed for the spectral image archiving recently. PCA, NMF and NTF were applied as a spectral reconstruction, a spectral reduction and feature extraction methods. It also supposes a multiresolution approach in computing NTF and subspace clustering preprocessing for compression by PCA. The experiments performed during the study shows that the non-negative methods reconstruct spectra with the same error but as the benefit they can be implemented optically. The compression method based on subspace clustering is more efficient than convenient k-means. The non-negative basis is better color feature than orthogonal one in a way of spectral image retrieval. Universal Decimal Classification: 004.93, 519.237.7, 512.643.12,514.743