With the advent of a single-pixel camera, compressed imaging has gained wide interests. However, the design of efficient measurement basis in such a system remains a challenging problem. In order to achieve fast and accurate compressed imaging, we propose a special family of measurement matrices, according to the coefficients distribution characteristics of natural images in sparse domain. The measurement matrix which can precisely locate the position of zero or approximate zero coefficients significantly outperforms the commonly used random matrix. And combining with the specific blocking strategy, the reconstruction speed can be increased linearly and the reconstruction performance is almost unchanged. Moreover, there is no blocking artifact. Several numerical experiments verify the validity of the measurement matrix. All image blocks are completely independent, so parallel fast compressed imaging could be realized.