Abstract: The features and distribution of one class of image largely represent their class information.In this paper, integrated with pixel information and shape information, we propose a method for image category recognition.The algorithm uses the following steps.First, given a specified image category, extract the salient features of each image, and then abstract the feature regions using the distribution of salient features.Second, aiming at get the feature dictionary, we cluster the feature regions.Based on feature words and shape information, we build the model, meanwhile we use the maximum likelihood estimation to learn the model parameters.Finally, by combing feature words model and shape model, test images are recognized.Experimental results show that our proposed method can categorize and recognize two classes of image effectively.The proposed method can also use for multiple classes of image.