In order to improve the efficiency of image retrieval of massive image database, an image retrieval method of massive data in the cloud computing based on the improved visual descriptor is proposed. Firstly, an extraction algorithm of visual features and an extraction algorithm of image interest points are designed; secondly, traditional binary-search tree are modified, the image ID of each interest point pair is set as the key of the tree, the search efficiency is improved by using binary-search tree; Finally, an index schema based on the key-value data structure is designed to organize small image files into a big index file, MapReduce of distributed cloud computing is adopted to compute the similarity of query image and image database. The experimental results based on the public high dimensional dataset and the big dataset show that, the proposed algorithm performs well in image retrieval accuracy rate, and improves the time efficiency of image retrieval of big data.