Hash method is an effective method for generating hash codes in large-scale image retrieval. The current hash method extracts the characteristics of the whole image first and then generates hash code, but the obtained hash code is not very precise to obtain more precise retrieval effect. Aiming at this problem, we propose a new method. First, we use a convolutional neural network to extract image features. Then, we adopt hash algorithm and generative adversarial network of input binary noise variable to learn image binary hash code, and carry out image similarity comparison by using hamming distance. Finally, we complete the effective retrieval of image data. Experiments on standard image data sets show that this method can effectively perform image retrieval, and the retrieval performance is improved than other methods.