Abstract: In AdaBoost algorithm, a lot of redundant features existed in the training of the classifier. In allusion to the problem mentioned above, an AdaBoost algorithm with integration of feature selection was proposed. Firstly, a feature selection method was developed for our model. Those visual features with small correlations between them was chosen to establish the optimal training set. Secondly, the AdaBoost integrated classification algorithm was used to train the classifier and to establish the classifier model. Finally, the classifier model mentioned above was used to realize the automatic annotation of the unlabeled images. Based the university of Washington image dataset, the experiment results showed that the proposed feature selection method was very suitable for the classifier, and compared with other classical algorithms, the algorithm given in this paper had the optimal accuracy classification.