An Alternative Approach for Pattern Detection Applied to Materials Characterization

Abstract

The problem of detecting specific patterns in images of materials obtained through High Resolution Transmission Electron Microscopy is addressed. A supervised classification method is proposed using an extension of Principal Component Analysis and a new a procedure for building the training set. Experiments on two different types of images indicate that the proposed method is superior to the conventional cross-correlation approach. Moreover, using the same number of components, the new dimensionality reduction approach shows a better performance than the standard PCA method.