Based on the analyses of current technology of SPCs and the compressed sensing theory, this thesis studies the bottleneck of the SPC, including the measurement matrix design within the exposure phase of imaging and the reconstruction algorithm within the signal reconstruction phase. The main contributions are: 1) Summarizes the functional structure, sampling principle and algorithms of the current SPCs, quantitatively explains why Hadamard matrices can be the measurement matrices of SPCs, and demonstrates the compressed sensing nature in SPCs by using the concept of the “total variation sparse domain”. 2) Studies the current dilemma of SPCs in the exposure phase (using hardware random number generator for measurement matrices has high performance in the exposure phase, but low efficiency in reconstruction because of the lack of a structural measurement; using permuted Hadamard matrices for measurement matrices has high efficiency in reconstruction, but this scheme requires unacceptably large memory to store the matrix data, and suffers a long loading time in the exposure phase), and proposes a hardware structure that can generate permuted Hadamard matrices benefiting from the advantages of both conventional methods. 3) Studies the time-consuming operations in the state-of-the-art algorithm for the image reconstruction of SPCs, and proposes a general parallel algorithm for the fast Hadamard transform. The study of the thesis presents a feasible fast exposure scheme for mega-pixel imaging of SPCs, which can be applied to other large-scale binary compressed sampling problems because of its simplicity and efficiency. Besides, since the fast Hadamard transform is widely used in diverse fields such as signal and image processing, communication systems, and digital logic, the proposed task-level parallelized Hadamard transform will have contributions for the corresponding large-scale problems in these areas.