A new method of content-based image retrieval which integrates color and texture features is presented. In order to improve image retrieval speed and accuracy,a new method is presented to show the mayor color,global color and texture feature of image using binary information feature,and two filters with these binary information are constructed to filter irrelevant images from image database. Then,similar measure is based on improved color histogram and new rotation complex wavelet feature. In order to improve retrieval performance,a relevance feedback mechanism is introduced. This feedback enhances the retrieval effectiveness by combining semi-supervised and active learning. Experiments show that the proposed system is not only superior to other methods but also effective.