Image synchronization sorting-retrieval optimization algorithm was proposed to solve the problems of low retrieval efficiency and accuracy of the existing algorithm. The image retrieval model was designed by introducing the concept of reverse learning. The complex lossless function was used to design the image retrieval mechanism and optimize the training error. Taking into account the attribute relevance of the query terms, we divided the training images into multiple subsets, and constructed the image ranking model by coupling the weighted factors to improve the effectiveness of attribute based image search. For given multi-attribute queries, the algorithm proposed in this paper could complete the retrieval using the word attribute contained in the queries. This algorithm supported multi-label queries, and had higher retrieval accuracy (when the recall rate was 80%, the accuracy was increased by 8.3% and 13.2% respectively as compared to the control group) and efficiency for multi-attribute queries as compared to the existing image sorting-retrieval mechanisms. In conclusion, the algorithm proposed could support multi-attribute queries and further increase the retrieval accuracy.