Compressed Sensing (CS) is a method of signal sampling that can directly acquire the compressed form of original signal. However, its signal reconstruction needs to solve the optimisation algorithm with high computational complexity. This paper introduces a redesigned algorithm of block CS, Block More Relaxed Regularised Orthogonal Matching Pursuit (BMROMP), which is a fast reconstruction algorithm for 2D-signals based on Regularised Orthogonal Matching Pursuit (ROMP) algorithm. To reduce the consumption of computational resources when reconstructing 2D-signals with big-size, BMROMP uses the method of dividing blocks reconstruction. For the reconstruction of each block signal, like ROMP, BMROMP also uses the least-squares method, but relaxes the calculation of the most relevant atoms. It is realised by a newly defined parameter, regularised-entire-correlation. The parameter can help us obtain 2D reconstructed signal directly. The experimental results show that BMROMP has great performance advantage in the reconstruction speed, i.e., it can drastically reduce the execution time of the optimisation algorithm. This point is very useful in many scenarios of CS application.