This paper presents a novel approach to the problem of obtaining a low dimensional representation of texture (pixel intensity) variation present in a training set after alignment using a Generalised Procrustes analysis. We extend the conventional analysis of training textures in the Active Appearance Models segmentation framework. This is accomplished by augmenting the model with an estimate of the covariance of the noise present in the training data. This results in a more compact model maximising the signal-to-noise ratio, thus favouring subspaces rich on signal, but low on noise. Differences in the methods are illustrated on a set of left cardiac ventricles obtained using magnetic resonance imaging.