WSEAS Transactions on Signal Processing
Print ISSN: 1790-5052, E-ISSN: 2224-3488
Volume 9, 2013
Decorrelation of Multispectral Images, Based on Hierarchical Adaptive PCA
Authors: ,
Abstract: In this work is presented one new approach for processing of groups of multispectral (MS) images, called Hierarchical Adaptive Principal Component Analysis (HAPCA). The aim is to decorrelate each group of N multispectral images, obtained after dividing the whole set into subgroups of 2 or 3 images each. In result, the basic part of the power of the images in one group is concentrated in a small number of eigen images only. This is achieved using the well-known method Principal Component Analysis (PCA) with transform matrix of size NN. In this case however, the method implementation needs high computational power, because it is based on iterative algorithms. Unlike it, the 2-level HAPCA permits to use transform matrices of size 33 (or 22), instead of the PCA transform matrix of size 99 (or 88 correspondingly), which makes the needed computational complexity 2 times lower in average. One more advantage of the new algorithm is that it permits parallel processing of each image sub-group in all hierarchical levels. In this work are also given some experimental results for the HAPCA algorithm applied on groups of MS images, which confirm the high decorrelation obtained. The proposed algorithm could be used as a basis for the creation of new algorithms for efficient compression of sets of MS and medical images and video sequences, for minimization of objects feature space in sequences of images, etc.
Search Articles
Keywords: Image processing, Image segmentation, Image contents analysis, Lossless image compression, Histogram modification, Inverse pyramid decomposition, Lossy image compression