To make the local robust feature-based approaches more effective and working on mobile devices, we propose using histogram of oriented gradients (HOG) features to characterize image regions. We propose an algorithm based on the entropy of HOG to select “good” regions for image matching and registration.Based on the characterization of regions, we can fine tune the feature selection and manage the feature distributions in different regions, so that image registration and matching can become much more efficient and at the same time more reliable.