Microwave nondestructive testing and evaluation (NDT&E) has tremendous potential for defect detection in metallic materials. In this paper: 1) an open-ended waveguide-based scanning system operating in the X-band (8.2-12.4 GHz) with a spatial-frequency feature extraction algorithm for defect detection at large lift-offs is presented; 2) a full mathematical derivation for modeling the spatial-frequency characteristics in the presence of defects and without defects is provided; and 3) a spatial-frequency feature extraction algorithm using the Itakura-Saito nonnegative matrix factorization is developed and investigated. The algorithm has the unique property of scale-invariance, which enables extraction of spatial-frequency features that are characterized by large dynamic ranges of energy. To evaluate the proposed technique, four defects in an aluminium plate with different depths (from 2 to 8 mm) and one tiny defect on a steel sample (0.45-mm width and 0.43-mm depth) have been examined. Experimental results have demonstrated that the proposed microwave NDT&E technique is capable of detecting defects at large lift-offs, with the potential of estimating the width and depth of defects, as well as classifying the different defect and nondefect areas.