Mechanical behavior of materials that contain domains of varying strength and ductility is commonly studied by subjecting them to standard tensile tests and measuring their full-field deformation using Digital Image Correlation (DIC). Though stereo DIC is a more appropriate tool than 2D DIC, limited studies have used high-resolution stereo DIC to measure heterogeneous strains at µm-length scale because it requires a sophisticated imaging equipment in addition to a more involved workflow than 2D DIC. Moreover, all these studies use the conventional stereo-DIC workflow, which has three problems that make these measurements unappealing and difficult to perform. They are: 1. temporal and stereo correlation involve an optimization problem that is sensitive to initial guess (initial-guess problem), 2. calibration is challenging since expensive miniature targets and precise translation stages are required (calibration problem), and 3. the underlying strains are smoothed when the conventional strain computation based on data fitting is used (strain-computation problem). In the present work, a novel version of stereo DIC called feature-assisted stereo DIC is developed by solving the aforementioned problems. Next, feature-assisted stereo DIC is applied to study the deformation heterogeneity at µm-length scale in two important applications: 1) tensile deformation of a copper oligocrystal and 2) tensile deformation of a bimetal prepared by fusing a Co-Cr alloy called Stellite on to a Ni-Cr-Mo alloy called Nicrofer using laser cladding. Initial-guess problem is solved by doing a limited but diverse search over the computer vision literature to find a feature-based matching technique to drive the DIC optimization. Calibration problem is solved by proposing a systematic cost-effective self-calibration procedure that makes use of speckled specimens under study itself as calibration target. Strain-computation problem is solved by using TVPCA, which is a strategy that combines total variation regularization and principal component analysis (PCA). Each of these improvements is individually validated on synthetic and experimental data. The improved stereo-DIC workflow is used to reveal the deformation heterogeneity in the two applications, and the heterogeneous strain maps are partially verified by conducting post-deformation microstructure analysis.