Learning and extracting primal-sketch features in a log-polar image representation

This paper presents a novel and more successful learning based approach to extracting low level features in a retina-like (log-polar) image representation. The low level features (edges, bars, blobs and ends) are based on Marr's primal sketch hypothesis for the human visual system. The feature extraction process used a neural network that learns examples of the features in a window of receptive fields of the image representation. An architecture designed to encode the feature's class, position, orientation and contrast has been proposed and tested. Success depended on the incorporation of a function to normalise the feature's orientation and a PCA pre-processing module to produce better separation in the feature space.