Abstract: Scale-invariant feature transform (SIFT) algorithm is one of the most dynamic algorithms which applied in image matching field.However,the computation complexity of descriptors is especially high.According to existing design method,the throughput rate of the whole module is only 50%.To conquer this bottleneck,we present a new sample method of descriptors in SIFT.The hardware cost of the structure is both small and worthwhile,since the sampling rate of the effective pixels is increased significantly.With 1.16% extra hardware cost,the average throughput reaches 77%.Compared with traditional method,the throughput improved 54%.