Segmented compressive sensing

This paper presents an alternative way of random sampling of signals/images in the framework of compressed sensing. In spite of usual random samplers which take p measurements from the input signal, the proposed method uses M different samplers each taking p_i´(i = 1, 2, 3 ... M) samples. Therefore, the overall number of samples will be q = M pmacr´. Using this method a variable sampling criterion based on the content of the segments is achievable. Following this idea, the calculated measurement (or sensing) matrix is also more incoherent in columns comparing to other conventional methods which is a desired feature. Our experiments show that the reconstructed signal using this method has a better SNR and is more robust compared to the systems using one sampler.