To support distributed video content analysis for video sensors in machine-to-machine networks, a reconfigurable stream processor for distributed smart cameras is proposed in this paper. A coarse-grained reconfigurable image stream processing architecture (CRISPA) with heterogeneous stream processing (HSP) and subword-level parallelism (SLP) is proposed to accelerate various algorithms for computer vision applications of smart-cameras. Implementation results show that the proposed design outperforms existing vision processors in many aspects: the on-chip memory size, power efficiency and area efficiency are 18.2 to 87.4 times, 4.5 to 12.5 times, and 3.8 to 25.5 times better than the state-of-the-art chips. Moreover, the programmability of the proposed design makes it capable of supporting many high-level computer vision algorithms in high specification.