Abstract: In order to recognize document images (DIs) more efficiently, this paper analyzes the blind point of the gray co-occurrence matrix based features in document image recognition, and presents novel features based on rectangle frame histogram (RFH) for united recognition of document images. Experiments show that the proposed method is superior to the GCM (gray co-occurrence matrix) based method in terms of recognition rate (95.96%) and false alarm rate (0.92%).