A partition algorithm method with sparse reconstruction was proposed to partition measurements making full use of the information of the extended targets. The grid sampling of the image plane was carried out by using diffraction limited optical system′s features,a sparse reconstruction model was set up with a “super complete dictionary”. Effective information in the amplitude of every pixel was extracted by sparse reconstruction, non-target measurement was eliminated by physical features. Partitioning measurements was realized by using the reconstructed sub-pixel-level target location and amplitude information. Simulation results indicate that the proposed method correctly partitions all the targets as separate measurements about 30s earlier than traditional distance-based partitioning method when signal noise ratio is 6 dB. Due to the full and effective use of target information, the proposed method outperforms the traditional distance-based partitioning methods in terms of the partitioning results′ accuracy, especially in the cases of poor signal noise ratio.