Abstract: In order to improve the efficiency and accuracy of massive information retrieval and mining in the network, a new method based on distributed processing technology is proposed. Using distributed retrieval system of MapReduce engine, to achieve mass image information measure of searching for similar; using the MapReduce retrieval by mining model mapping and reduction operation process, can the image similarity metric operations, and the feature met with image merging; using sift and hog and color moments of the three characteristic parameters of image feature vector construct 3D visualization and image features corresponding to visual and semantic vocabulary, achieve the retrieval result display. The simulation results show that the improved algorithm is time-consuming, high accuracy and high accuracy.