The limited size of publicly available sparse matrix datasets creates a significant challenge for benchmarking, testing, and validating algorithms in scientific computing, artificial intelligence and other data-intensive applications. Existing approaches such as random matrix generators or general data augmentation methods often fail to produce structurally realistic matrices. To address this gap, we present MatGen which a tool for generating realistic variations of a given sparse matrix using signal processing and image processing techniques. MatGen takes a real sparse matrix as input and produces structurally consistent matrices at different sizes, introducing controlled variation while preserving key sparsity patterns. We evaluate the effectiveness of MatGen by analyzing structural features and visual similarities between original and generated matrices. Experimental results show that MatGen can produce realistic, scalable sparse matrices suitable for a wide range of applications including benchmarking computational methods, and sparse data techniques.