Principal component analyses (PCA) has been widely used in reduction of the dimensionality of datasets, classification, feature extraction, etc. It has been combined with many other algorithms such as EM (expectation-maximization), ANN (artificial neural network), probabilistic models, statistic analyses, etc., and has its own development such as MPCA (moving PCA), MS-PCA (multi-scale PCA), etc.

PCA — and its derivatives — has a wide range of applications, from face detection, to change analysis. Change detection from PCA shows however a main difficulty, that is, result interpretation. In this paper, a new PCA method is developed, namely MB-PCA (Multi-Block PCA), in order to overcome this problem. Experimental results demonstrate the interest of the approach as a new way to use PCA.