This document discusses information ratio (IR) and mutual information ratio (MIR) as methods for estimating the number of image features and feature matches. IR is defined based on image channel histograms and self-information. MIR similarly uses joint histograms and mutual information. Lower bounds on IR (LIR) and MIR (LMIR) are also proposed based on entropy. Numerical experiments evaluate IR and MIR on standard datasets using SURF, KAZE and ORB features. Results show these features follow the IR curve and extract fewer features than estimated by IR and LIR. Optimization of feature extraction based on IR is also shown to outperform standard algorithms. Future work is proposed to further evaluate M