Abstract: How to get a better effect for the image style transfer is one of the typical problems in the field of image processing. Aiming at the problems of poor performance and strong limitation in the traditional methods，this paper firstly discussed on the characteristics of image features extraction in the different layers of CNN（convolutional neural networks），then an image style transfer algorithm based on Gram matrix and CNN was proposed，and a variety of art style transfer experiments were designed. The experimental results showed that the proposed algorithm could better implement the style transfer of art images in a shorter time than traditional algorithms，and the experiments verified the advantages of the proposed algorithm in image feature extraction and style transfer tasks.

Key words: image style transfer, convolutional neural network （CNN）, Gram matrix, dilated convolutions