<p>With the expanding applications of wireless multimedia sensor networks (WMSNs), there is a growing demand for compression techniques that are more suitable than the traditional methods. Conventional methods involve high-rate sampling followed by compression algorithms. However, such techniques are not well-suited for low-power imaging devices with limited computational capabilities. Compressed sensing has offered a new paradigm in signal acquisition, in which a signal is reconstructed from few linear measurements, reducing the encoding complexity. However, decoding still suffers from excessive reconstruction time and limited reconstruction accuracy. In this paper, we propose block fast matching pursuit (BFMP) algorithm, in which the image is processed in blocks, where the 2D sparse domain is applied to each block, ahead of vectorization. Image reconstruction is performed iteratively selecting a variable number of atoms per iteration, and refining the support in each iteration. Our simulations have shown a significant increase in speed compared to existing methods at very high reconstruction accuracy. Furthermore, our approach is suitable for the cases of large images and noisy measurements.</p>