Scene-based Fourier pattern-sorting approach for compressive sensing

In defense and security applications cost-intensive high-performance imaging sensors are common assets. Modern approaches in the field of Computational Imaging promise the cost-efficient construction of high-performance imaging sensors with additional features, like multitask or hyperspectral imaging capabilities. For example, the compressive single pixel imaging method we studied benefits from a low-cost architecture and multimodal sensor capabilities. The compressive single pixel imaging approach enables the adaption of resolution and compression at the cost of operational speed. The measurement process of an imaging system based on this approach involves a spatial modulation of the scene with a spatial light modulator utilizing different modulation patterns as well as a subsequent reconstruction step to obtain an image of the scene. The computational speed of the image reconstruction is still a problem, especially for the class of iterative optimization algorithms. For real-time applications like threat detection a fast image data acquisition and reconstruction are fundamental though. The fast reconstruction of the images via a fast and efficient transformation, e.g. the fast Fourier Transformation, is advantageous when using binary transformation-based modulation patterns. Therefore, we studied binary Fourier patterns and compared scene-based evolutionary and other sorting methods for transformation based single pixel imaging in order to maximize the compression while controlling the image degradation regarding relevant features to enable real-time applicability.