This thesis primarily focuses on how to carry out intelligent sensing and understand the high-dimensional and low-quality visual information. After exploring the inherent structures of the visual data, it proposes a number of computational models covering an extensive range of mathematical topics, including compressive sensing, graph theory, probabilistic learning and information theory. These computational models are also applied to address a number of real-world problems including biometric recognition, stereo signal reconstruction, natural scene parsing, and SAR image processing.