Compressive sensing (CS) is a new sampling theory which allows reconstructing signals using sub-Nyquist measurements. It states that a signal can be recovered exactly from randomly undersampled data points if the signal exhibits sparsity in some transform domain (wavelet, Fourier, etc). Instead of measuring it uniformly in a local scheme, signal is correlated with a series of sensing waveforms. These waveforms are so called sensing matrix or measurement matrix. Every measurement is a linear combination of randomly picked signal components. By applying a nonlinear convex optimization algorithm, the original can be recovered. Therefore, signal acquisition and compression are realized simultaneously and the amount of information to be processed is considerably reduced. Due to its unique sensing and reconstruction mechanism, CS creates a new situation in signal acquisition hardware design as well as software development, to handle the increasing pressure on imaging sensors for sensing modalities beyond visible (ultraviolet, infrared, terahertz etc.) and algorithms to accommodate demands for higher-dimensional datasets (hyperspectral or video data cubes). The combination of CS with traditional optical imaging extends the capabilities and also improves the performance of existing equipments and systems. Our research work is focused on the direct application of compressive sensing for imaging in both 2D and 3D cases, such as infrared imaging, hyperspectral imaging and sum frequency generation microscopy. Data acquisition and compression are combined into one step. The computational complexity is passed to the receiving end, which always contains sufficient computer processing power. The sensing stage requirement is pushed to the simplest and cheapest level. In short, simple optical engine structure, robust measuring method and high speed acquisition make compressive sensing-based imaging system a strong competitor to the traditional one. These applications have and will benefit our lives in a deeper and wider way.