In the traditional image retrieval model,the image future is usually extracted from a fixed scale image, which will inevitably degrade the performance of the whole system.M otivated by this,this paper introduces a hierarchical sparse coding model to realize image retrieval i n multi-scale cues,and proposes a fast image retrieval technology with hierarchical matching pursuit.I n this method,the low-layer sparse codes features extracted from the image are transferred to the high- layend then the extracted high dimensional sparse codes featues are transformed into the improved PCAH feature.After that,the Hamming distance is used for fast image retrieval.It can be seen from the experiments o n two benchmark datasets of Caltech256and Corel5K that compared with other hash methods,our method improv es the precision and recall rate by 5% and 10% respectively,with the least running time.Experimental results demonstrate that the proposed method not only obtains better accuracy,but also has higher time effic iency.