Material characterization using electron backscatter diffraction (EBSD) requires indexing the orientation of each scanned pixel based on a Kikuchi pattern. Several approaches to indexing are available like Hough Indexing (HI) which transforms the pattern to a Hough space and Dictionary Indexing (DI), where a library of simulated Kikuchi patterns is compared to the test pattern revealing a closest match. The advent of artificial intelligence and data driven material science has further led to the development of trained machine learning models which are shown to be very effective in indexation and segmentation. However, such methods are computationally expensive and require large datasets and extensive training time. Furthermore, robustness remains an issue as models trained for one type of dataset are not easily transferable to other.

An effective way to circumvent these issues is the use of unsupervised data-based methods, which are easy to use and robust to deploy. There is a lot of scope in applying such methods, in innovative ways, to characterize microstructures. Principal component analysis (PCA) and non-negative matrix factorization (NMF) have been used to denoise and segment a microstructure effectively. In this work we showcase the latest data-based methods used for microstructure segmentation as well as solving specific tasks, such as pattern overlaps at grain boundaries.  We describe an application of a constrained non-negative matrix factorization scheme to segment low angle grain boundaries (LAGB) effectively, which is on par with high resolution EBSD techniques. This approach allows to resolve the location of a grain boundary at the pixel level. Finally, we introduce the concept of Embeddings-EBSD and showcase its applications in segmentation and detecting accumulation of localized strains. Together with the two approaches we present some illustrative applications in material characterization.