In this paper, a new method for image feature extraction is proposed based on the image classification recognition problem. Firstly defining the complexity of the image string and the universal image distance (UID), and then proposing a UID distance measurement algorithm to measure the general image distance, a prototype selection algorithm for selecting the image prototype under the inherent difference between the maintenance feature categories, a feature vector generation algorithm to generate the eigenvector of the image to be classified by using the prototype selection algorithm to create the feature vector of the image to be classified, and finally an image classification learning algorithm to separate the region of interest of the image based on the aforesaid algorithms. The proposed method is applied to several supervised and unsupervised learning experiments of satellite image data. The results show the feasibility of the proposed method.