Massive electronic data involve lots of image evidence. The specific classification of geometric curved surfaces on the image set in image recognition algorithm will result in information loss. For this problem, an improved iterative deep learning algorithm (IIDLA) based on convolutional neural network is proposed. The algorithm adopts pooled convolutional layer (PCL) to conduct underlying translation invariant feature learning, applies hierarchical iteration and uses convolutional neural network (CNN) to learn different nonlinear features of input image set. The image library and inquiry examples of the algorithm include face or object image set of different view angle, background, facial expression, resolution and illuminance. The dataset is used to evaluate and compare the proposed algorithm and other algorithms. The results show that the proposed algorithm has the best performance on the evaluation dataset.